praisethefool's picture
Add SetFit model
1beb54a verified
metadata
tags:
  - setfit
  - sentence-transformers
  - text-classification
  - generated_from_setfit_trainer
widget:
  - text: >-
      Untangle Solving problems with fuzzy constraints Untangle Solving problems
      with fuzzy constraints Szymon Kaliski Marcel Goethals Mike Kluev January
      2023 Have you ever needed to find a time for all your friends to meet for
      dinner or to create a seating plan for wedding guests These are examples
      of problems that require navigating a set of overlapping constraints you
      are only available on every other Tuesday or Thursday Chris will not show
      up if Ash is not there but Ash is in town only the last week of November
      and you really wanted to catch up with Chris We often work out problems
      like this with a pencil and paper experimenting until we find a solution
      but it feels like computers should be the perfect tool to help us In fact
      there are programming tools called theorem provers which are designed to
      solve exactly this class of problems They excel at helping experts work
      through fullyspecified problems with a clear solution Unfortunately we
      rarely have a formal understanding of our problems We come to understand
      them interactively by trying to find a solution In fact we often just need
      a solution that is good enough instead of one proven to be optimal We set
      out to experiment with interactive computerassistance for this type of
      thinking Untangle is a research prototype that helps you think through
      everyday constraint problems using your tablet stylus With Untangle you
      leave handdrawn marks on a page sketch out the representation of your
      problem introduce constraints graphically and browse through sets of
      possible solutions This work was presented as a part of our Programmable
      Ink talk at Strange Loop 2022 We welcome your feedback inkandswitch or
      helloinkandswitchcom Contents Thinking fuzziness and constraints Design
      principles Inspirations Logic programming Pattern matching Graphical
      production systems Realworld applications of logic programming Untangle
      Assignment problem Frequency assignment bids Coffee shop schedule
      Generative art Recursive rewrites Findings A smooth ramp from concrete to
      abstract helps form intuitions about the system Informal solving is most
      useful for certain kinds of problems Many problem representations look
      like tables Handdrawn input is well aligned with exploratory
      problemsolving Fuzziness and live feedback contribute to a conversational
      feel Shortcomings The system lacks clear semantics Turning ink into
      symbols is an unnecessary technical crutch Some parts of the system lack
      visibility Responses from the computer are underdesigned Small canvas
      artificially constrains the problem representation There is no way to
      explore the solution space Conclusions Thinking fuzziness and constraints
      Computers can be great at solving logic problems like the ones mentioned
      above as long as we can describe them in a formally correct and detailed
      way Special programming languages and techniques—theorem provers—exist and
      can calculate solutions for huge datasets These languages are most often
      used by mathematicians to help with proving formal theorems or by domain
      experts to aid in modeling large scale industrial production systems A
      Constraint satisfaction problem can be solved using logic programming
      techniques such as Satisfiability Modulo Theories SMT or Answer Set
      Programming ASP These programming languages are useful after we have
      encoded a problem in machinereadable form but first we must do the harder
      part fully understand the problem For this we often reach for pen and
      paper which allows us to think fuzzily and omit various levels of detail
      when problemsolving We can quickly sketch out the representation of the
      problem without worrying about absolute correctness For a sampling of
      realworld pen paper constraint problem representations have a look at How
      People Visually Represent Discrete Constraint Problems by Xu Zhu et al
      This project explores what it might look like if computers could support
      this style of earlystage thinking Untangle is specifically not a tool for
      solving artificial logic puzzles nor is it a tool for creating formal
      specifications for industrial systems Instead we are interested in a tool
      that can help us think through illdefined problems understand compromises
      and learn about what kind of questions to ask Untangle is a continuation
      of the threads highlighted in the thinking modeling and computers section
      of the Crosscut essay Mainly we want a tool in which we can sketch a
      dynamic representation of the problem at hand and have a conversation with
      it Design principles Keep focus on the problem not the implementationWe
      want a tool in which your focus remains on the problem at hand as much as
      possible rather than thinking about the correct way to encode it in a
      machinereadable way No errors undefined values or unknown parameters to
      fill inThe tool should never block freeze or become unresponsive even if
      the user creates invalid states such as errors or incomplete input A wrong
      answer is better than no answer Everything is visibleBoth the domain model
      and the constraints should always be visible and interactive Conversation
      with the material is be encouragedWe want an iterative approach to problem
      solving—one where observing leads to thinking which leads to acting which
      leads back to observing You should be able to intuit connections between
      various rules and constraints by wiggling them and seeing other things
      wobble We also adopted most of the design principles from Crosscut A
      tablet stylus can become dynamic pen paper The content of what you are
      working on is the most important thing You should not have to use an
      onscreen keyboard for programming This is a personal thinking space
      Untangle shares a lot of context with Crosscut a research project in which
      we explored an approach to building dynamic models by direct manipulation
      There is one important design difference from Crosscut using handdrawn
      strokes instead of vector graphics We believe there is something special
      about leaving distinctively human marks on the page with a stylus so we
      want to go back to handdrawn marks like with Inkbase but with an entirely
      different computational model Inspirations Logic programming Logic
      programming is an important paradigm in computer science It is based on
      formal logic and allows programmers to describe problems using declarative
      statements “every human is mortal” and ask questions based on these
      statements “X is human is X mortal” Some notable logic programming
      languages include Prolog one of the first logic programming languages
      widely used for theorem proving term rewriting automated planning Z3
      Theorem Prover a satisfiability modulo theories SMT solver that is
      targeted at software verification and program analysis Untangle uses Z3 as
      a library for solving Alloy Analyzer Alloy is a language for describing
      and exploring software models It has been used in a wide range of
      applications from finding holes in security mechanisms to designing
      telephone switching networks Of particular interest is Alloy’s IDE that
      visualizes a possible structure based on the constraints provided by the
      user Alloy Analyzer’s interactive solver visualization Pattern matching
      Untangle relies heavily on spatial queries—finding symbols on the page by
      their spatial relation to other known symbols—which were inspired by Regex
      a domainspecific language for describing search patterns in text
      Qualitative Spatial Reasoning a calculus which allows a machine to
      represent and reason about spatial entities without resorting to
      traditional quantitative techniques QSR is often used in GIS Geographic
      Information Systems for querying geographical data For more information on
      QSR check out A survey of qualitative spatial representations by Chen et
      al Graphical production systems Work on Untangle was also inspired by
      graphical production systems which use shape matching rules and graphical
      rewrites to describe computations Publications that guided our work Shape
      Grammars and the Generative Specification of Painting and Sculpture a
      seminal paper by Stiny and Gips introducing shape grammars New graphical
      reasoning models for understanding graphical interfaces a paper from
      Furnas introducing BICPICT a pixelrewriting graphical reasoning system
      Wave Function Collapse an approach for generating tile maps from a single
      example which influenced our thinking on using superposition as a mental
      model for working with multiple possible values Wave Function Collapse
      algorithm visualzation Figure by Maxim Gumin Realworld applications of
      logic programming Finally we were guided by various examples of using
      theorem provers for working through everyday problems How people visually
      represent discrete constraint problems Using linear programming GLPK for
      scheduling problems Tax planning with Z3 Theorem Prover A shared pinboard
      becomes a collaborative modeling tool to plan a dinner party Untangle We
      are now going to introduce Untangle a tool for working out realworld logic
      problems Assignment problem To explore the basic concepts of the tool we
      are going to look at an assignment problem imagine you are teaching a
      class it is the end of the semester and each student needs to submit a
      short paper about some topic You want students to grade each other so you
      have less work to do Let us see how Untangle can help us solve this
      problem This reallife problem comes from Joshua Horowitz thank you Josh In
      its most basic form Untangle allows you to automatically assign symbols to
      other symbols For example you can put Bob into a box by drawing an arrow
      between the text Bob and the box—Bob will now appear in pink inside the
      box A symbol is anything that appears on the canvas Symbols can be a
      single stroke or collections of strokes Assigning Bob to a box We can also
      assign multiple symbols to a box Untangle will now show that there are two
      possible assignments by showing two dots at the bottom of the screen We
      can scroll through the different solutions Bob and Eve Assigning multiple
      students to the same box There are six students in our class so we will
      write all of their names Instead of drawing six arrows we can use spatial
      queries to find elements on the canvas Using a spatial query to grab all
      of the student names Whenever we make a selection of symbols on the canvas
      a popup appears This popup suggests different spatial queries we can use
      to match our selection In this case it suggests that we could look for a
      vertical column of symbols starting with Bob Representation of a spatial
      query capturing a column of symbols starting with Bob We can place this
      query onto the canvas and use it as a shortcut to refer to all the symbols
      that match it The matching symbols are highlighted in the same color for
      additional visual feedback Untangle only ever places one symbol into each
      box Whenever there are multiple ways to assign symbols Untangle will
      generate multiple solutions but display only one The dots at the bottom of
      the screen indicate that there are alternative possible assignments We can
      get a sense of the different solutions by scrolling through them Solution
      switcher We want to assign a student to every other student so let us
      create an empty box next to every name Just like with the spatial query
      for a list of names we can create a query that simply looks for all boxes
      This generates all possible ways to assign one student to another one
      Using the all boxes query to assign every student to another student
      Currently students are sometimes assigned to grade themselves While
      students will surely be happy to do so as a teacher I would rather avoid
      this situation Let us add a rule that prevents students from grading
      themselves To do this we will add a third spatial query This query matches
      any box that has something to the left of it The question mark acts as a
      wildcard matching any symbol Using wildcard to capture something next to a
      box and make that something not equal to the contents of the box Finally
      we can draw an inequality constraint arrow This expresses that whatever
      ends up in the box cannot be the same as whatever symbol is on its left
      Bob can no longer grade Bob Inequality constraint Let us say we omitted a
      student when writing the original list of names Because we are using
      spatial queries to describe the column of student names we can easily
      extend the list Extending the student list with a new name The results
      update reactively It turns out that Bob and Claire have written papers
      about similar topics So it would be great if they grade each other We can
      ensure this by simply putting their names into the corresponding boxes and
      the system will adapt accordingly Forcing the assignment of Bob to Claire
      Again the system reacts with new results Frequency assignment bids To show
      a few more interesting properties of Untangle let us look at a different
      example—running frequency spectrum assignment bids The basic idea is that
      a specific radio frequency spectrum say 5G is divided into smaller parts
      and sold to various operators This reallife problem comes from William
      Taysom thank you William Let us start by modeling this problem There are
      three telephone companies telcos that are bidding on the frequency
      spectrum from 34 GHz to 42 GHz The spectrum is split into eight bands and
      the telcos can bid on individual bands Initial model of the spectrum
      assignment problem Just as in the student grading example we could use
      spatial queries to assign telcos randomly to bands in the spectrum
      Assigning operators to random bands of the available spectrum But this is
      not really how an auction works Instead telcos bid on a specific number of
      bands that they want to obtain For example Verizon might place a bid for
      four bands We can model this using a count modifier which limits the
      number of times a certain rule applies Using a count modifier to limit the
      assignment to four times Let us rearrange the canvas a bit and add some
      imaginary bids for each of the companies Modeling bids from all of the
      operators Some bands can be more valuable for example the middle bands
      often have less interference so a company may bit not just on a specific
      number of bands but also specific placement inside the spectrum To model
      this we can either draw an arrow directly to the specified band or simply
      drag a symbol in place We can guide the solver into a direction that we
      care about and it will respond immediately Constraining the solutionspace
      further by modeling bids on specific bands in the available spectrum The
      number of bids might not equal the number of available slots—an
      overconstrained system Using most solvers the result would be an error
      message that the constraints are unsatisfiable Instead of showing an error
      Untangle will attempt to generate a partially correct solution by ignoring
      some of the rules Arrows will turn red indicating that for the currently
      shown solution this rule is ignored Relaxing the overconstrained problem
      representation by ignoring some of the rules When having a conversation
      with the material hearing “no but what if…” is more encouraging than
      hearing just “no” The machine does not brake or scold at you for making a
      mistake but instead shows compromises and possible directions which in
      turn helps generate new ideas Finally companies sometimes bid specifically
      on a set of consecutive bands—rather than just bidding on four bands they
      want four bands in a row We can model this using a combination of spatial
      queries and counts Creating an assignment rule of four consecutive bands
      using a combination of spatial queries and count modifiers Coffee shop
      schedule In the following example we will show how to create a work
      schedule for a coffee shop An easy solution is to assign employees to
      random days but to make a good schedule everyone is happy with we need to
      consider employee availability To solve this problem we need to two
      dimensions First we list availability of the employees in the topleft
      corner For each day that an employee is available we draw an X For example
      Jim is available on Monday Wednesday and Friday In the bottom right we
      draw a simple empty schedule for the week Modeling the baristas
      availability Untangle has a special lookup spatial query that allows us to
      look up information in tablelike layouts In this case it matches columnrow
      pairs together whenever it finds an X Using the lookup spatial query on
      the availability table You can think of this as finding all the available
      people for each day Untangle will only generate solutions where the day
      and name are matched in the availability table Visualization of running
      the lookup spatial query We then create a spatial query to find all of the
      empty slots in the schedule using a something on left box on right query
      Spatial query using a wildcard symbol and a box to the right of it We only
      want to assign employees to days they are actually available To do this we
      bind the wildcard matches on days of the week We then use the names to
      fill in the boxes in the schedules Untangle will show us all possible
      schedules based employee availability Resulting coffee shop work schedules
      based on employee availability The proposed schedules often have people
      working two days in a row—something we might like to avoid We can set up a
      simple constraint to make sure no two boxes in a row hold the same value
      Constraining consecutive not to hold the same value It is worth noting
      that in this example technically all the information needed to solve the
      problem is already on the canvas However coming up with a valid solution
      still requires serious “System 2” level thinking shuffling around symbols
      in your head It feels great to have the system do this part for you System
      1 System 2 is a dichotomy introduced by Daniel Kahneman in Thinking Fast
      and Slow System 1 describes fast intuitive loweffort thinking while System
      2 is effortful and slow Generative art So far all the examples we looked
      at were about constraining the solution space—progressively going towards
      a small set of satisfying solutions This is only a one part of the
      problemsolving process which has two distinctive phases that feedback into
      each other—expanding and collapsing Expand and discover different
      possibilities then narrow scope and focus We can use Untangle’s primitives
      to force us to expand the possible solution space instead Let us
      illustrate this by recreating one of the most famous computer art pieces
      10 PRINT 10 PRINT is a oneline Commodore 64 BASIC program that generates a
      visual pattern on the screen We can make a couple of boxes two symbols and
      fill all boxes Using spatial queries to fill all of the boxes with
      permutations of diagonal lines This already starts to look interesting but
      we can get to more compelling results by adding additional constraints
      Forcing more interesting results by applying inequality constraints And of
      course we can introduce additional symbols and keep exploring the solution
      space 10 PRINT variation using three symbols and two inequality
      constraints Recursive rewrites We can apply the rewrites recursively by
      flattening the pink results back onto the canvas and turning them into
      black ink This will in turn update the spatial queries and generate a new
      set of results New graphical reasoning models for understanding graphical
      interfaces is a seminal paper on using graphical rewrite rules for
      computation Progressively flattening solver results back onto the canvas
      to create basic 1D cellular automata Recursive rewrites can be used in a
      lot of interesting ways As an example below is a recreation of a logic
      gates demo from Inkbase We start by creating the symbols And the rules
      they follow Which shows us immediate feedback on the canvas We can then
      set up some logic gate networks and propagate the values through them by
      writing them onto the canvas to get to the final result A process of
      propagating values through two logic gate networks Findings We have used
      Untangle to solve the problems outlined above as we were building the
      prototype to test our assumptions and intuitions We also conducted several
      informal interviews with potential users with background in mathematics
      and logic programming Here are some reflections from that process A smooth
      ramp from concrete to abstract helps form intuitions about the system
      Symbols queries and arrows build up on each other Arrows are a reified way
      of moving a symbol into a box manually Spatial queries are a reified way
      of selecting things manually Combining the two is a reified way of drawing
      multiple arrows between multiple symbols “Reification” means giving a
      concrete representation to an abstract process In this sense each
      primitive is simply a way of expressing in more general terms what you
      could already do in an earlier step Gradually climbing this ladder of
      abstraction helps build intuition about how the system behaves Informal
      solving is most useful for certain kinds of problems There seem to be two
      dimensions of complexity for a given problem a problem can have relatively
      trivial constraints but many elements that need to be solved or can have a
      small number of elements but constraints that are difficult to satisfy
      There is a sweet spot where Untangle seems most useful The dataset must be
      small enough to be manually drawn but with constraints too complex to
      easily solve in your head If the dataset is large it takes too much effort
      to write it all down and the tablet screensize limits what can fit on the
      canvas If the set of constraints is simple enough to keep in your working
      memory you can often just solve the problem as you create its
      representation without additional help from the computer We found that the
      even a small number of overlapping constraints forces us to switch to
      effortful “System 2” thinking and this is where Untangle shines helping us
      think through a problem in an informal way Many problem representations
      look like tables The way we naturally represent assignment problems tends
      to drift toward using tables Examples of tablelike structures invented
      adhoc to represent specific problems Even though a structured
      approach—like a builtin table tool—might seem more appropriate to modeling
      these kinds of problems freeform input feels important The exact shape of
      the problem representation might not be clear initially and sketchiness
      facilitates finding it It feels good to build up to a table rather than
      being prematurely forced into one Another possibility is that the provided
      spatial queries column with symbol at top row with symbol at left table
      lookup etc encourage drawing problems in gridlike structures A rich area
      for future work would be adding more ways to query the canvas which could
      lead to more diverse representations Handdrawn input is well aligned with
      exploratory problemsolving One of our motivations for handdrawn input was
      to enable drawing symbols and elements that mapped closely to a particular
      problem domain For example you could draw chairs and tables for a seating
      arrangement In practice we rarely drew domain objects instead favoring
      symbols like names logos or even dashes and dots However handdrawn input
      still felt much more aesthetically fitting than the vector version we
      tried early on Early iteration of Untangle using vector graphics and
      artificial symbols At the lab we believe that the fidelity of the tool you
      use should be proportional to the maturity of the idea you are working on
      Being forced into crisp vector shapes for exploratory problems creates a
      cognitive dissonance between the fuzzy nascent problem in our head and the
      precise symbols on the canvas Fuzziness and live feedback contribute to a
      conversational feel Untangle was designed to get the user to a result as
      fast as possible Simply drawing a single arrow can generate multiple
      results To find a satisfying solution it is often not even necessary to
      add additional constraints Instead you can just scrub through proposed
      solutions to find one that makes sense If you find a satisfying solution
      you can simply stop working even if the results are underspecified In a
      similar vein if the solver cannot satisfy your constraints rather than
      showing an error message the system will attempt to ignore some
      constraints relaxing the problem statement to generate a technically
      incorrect solution The interface will highlight the arrows that were
      ignored to generate each solution In realworld contexts we often are not
      trying to find the globally optimal solution but rather just any
      reasonable one Spatial queries are also fuzzy We do not look for something
      “exactly 137px to the left” but “roughly to the left” This plays well with
      handdrawn aesthetics as you never create perfect sketches Additionally
      spatial queries highlight their matches directly on the canvas This has
      two advantages first it helps explain what the query is doing—even if you
      have no idea of the underlying formalism you can experimentally find out
      what is happening Secondly it makes it transparent when the imperfect
      matching algorithm does not work as expected You can always just wiggle
      your drawing a bit to get the system to recognize it Finally arrows point
      to and inside of queries rather than of connecting to ports This has a
      specific informal feel which meshes well with the fuzziness of other parts
      of the system and is distinctively different from the feel of things
      snapping into each other Shortcomings The system lacks clear semantics
      Untangle’s “language” is unspecified and not very composable The set of
      provided spatial queries is adhoc We created new queries for problems we
      were solving in the examples instead of building them up from first
      principles As a result we can solve the examples nicely queries are not
      composable or abstractable—you cannot combine to the left and to the top
      to create the lookup query used in the coffee shop schedule nor can you
      push the results from one query into another one to create reusable
      functionalities Additionally creating these queries purely by example can
      be quite tedious—especially when it comes to selecting wildcard
      configurations It seems clear it would be better to create a match by
      example then interactively refine it through direct manipulation One
      exciting piece of research in this direction is described in Perceptual
      Grouping Selection Assistance for Digital Sketching by David Lindlbauer et
      al Most importantly it is hard to intuit how the primitives behave beyond
      basic demos For example there is a subtle difference between constraining
      elements of the solution space and constraining elements of the match
      Careful thought is required to discern the difference Turning ink into
      symbols is an unnecessary technical crutch Untangle is a system for
      assigning candidate symbols to potential targets In our experience these
      symbols often consisted of many ink strokes such as for people’s names In
      order to assign these to a target we need to recognize those strokes as
      being part of a grouping Using a magic wand to group multiple strokes into
      a single symbol In past research we found that it was very difficult to
      reliably and unambiguously group ink strokes or to recognize repetitions
      For this project we simply sidestepped the problem with a command to
      create a symbol out of a group of strokes This was convenient for a
      research prototype but we feel this is unfortunate technical ceremony
      Similarly when creating a drawing we encourage users to reuse copies of
      the same target symbol to hint the solver system that those targets are
      related Sometimes copying objects can be a fast and intuitive thing to do
      Other times—especially when the shapes are simple—it feels more natural
      just to draw them again and have them match automatically The history of
      inferring specific user intent from ink gestures remains an open problem
      after many years of related work including work on shape recognition for
      drawing programs and back to early stylusbased text input systems such as
      Palm’s Graffiti Some parts of the system lack visibility A computing
      system should never leave a user feeling uncertain about whether the
      intent of the user has been understood In Untangle assignment arrows are
      freeform they can exist with or without valid sources and targets This
      meshes well with the fuzzy aesthetic but our implementation does not
      provide user feedback whether the arrow has actually connected the items
      on the canvas other than the updated solutions For this specific example
      it is easy to imagine how to improve this For example by color coding
      arrows and connected spatial queries the same way we do for query matches
      However color coding query components as well as their matches creased a
      visually noisy rainbow canvas Finding the right queues and feedback for a
      system like this is a subtle task There is a fundamental tension between
      maintaining a focus on the user’s input by avoiding unnecessary UI chrome
      and preventing confusion by providing sufficient feedback Responses from
      the computer are underdesigned In systems where users and computers
      collaborate it helps to distinguish between user input and computed
      responses In Untangle user input is always black and we use pink to
      distinguish computed results from userdrawn strokes We refer to this as
      the user voice vs the computer voice Throughout our research we often want
      the machine to respond and draw with us but what is the “correct” way to
      do so Because Untangle limits user input to a single color rendering
      results in pink makes the distinction immediately clear but we worry this
      would become problematic in a more fullfledged system We briefly explored
      rendering computed results using a different “pen” for example a stylized
      marker but that felt uncanny—the user’s input strokes redrawn exactly but
      with a different aesthetic Small canvas artificially constrains the
      problem representation In Untangle the relatively small screen size and
      lack of support for canvas features such as panning or zooming limits the
      amount of data and the complexity of the problems you can represent
      Drawing everything by hand also contributes to this limitation—it simply
      requires too much effort to draw hundreds of student names or create a
      staff calendar for an entire year We are interested in how the experience
      of Untangle would evolve as we explored larger scale problems or more
      complicated representations For example one future project we would love
      to see is “Untangle with external data” There is no way to explore the
      solution space One omission in this work is the limited ability to
      visualize the solution space Yes we can narrow down the solution space by
      reifying a result “Sam has to review Steve’s paper” or by adding
      constraints “the wildcard cannot be the same as the contents of the box
      next to it” but while Untangle allows you to “scrub” through results it
      only ever shows one result at a time In fact the solution space is not a
      homogeneous list there are recurring patterns It would be interesting to
      explore visualizations that revealed clusters or branches of candidate
      solutions that share similarities or how how different constraints “cut
      off” certain areas of the solution space Conclusions With this project we
      set out to discover what a nonbureaucratic theorem prover might look like
      The traditional programming interface to a theorem prover is both strict
      and formal Untangle shows a glimpse of a computational model with
      fuzziness at its core Being able to handwave at a problem and get to
      results—often on the first browse through the solution space—feels
      wonderful and is a stark contrast Spatial queries provide a way to create
      structure on top of a freeform drawing Instead of forcing problems
      prematurely into tabular form you can start sketching the problem however
      feels natural You then work with the system to query that diagram for a
      satisfying solution In some cases the final result may take the form of
      more traditional tabular data but we found that building up to it from a
      freeform drawing and not being constrained by it prematurely allowed us to
      explore our ideas more naturally The combination of symbols spatial
      queries and arrows provides a nice onramp for abstracting logic Rules can
      be built up out of simple examples gradually adding assignment arrows or
      replacing those arrows’ concrete sources and targets with spatial queries
      We feel this conceptual buildup is very promising and points at a possible
      way of solving the repetition problem described in the Crosscut essay
      Untangle is part of our “programmable ink” track of research continuing
      from previous projects Inkbase and Crosscut We remain optimistic about
      systems in which you directly manipulate the representation of the problem
      at hand and that remain alive and reactive This combination allows you to
      improvise and rely on intuitions instead of having to switch your thinking
      mode to one of effortful logical computation We see here an exciting
      glimpse of conversation with a dynamic medium—sketching at the speed of
      thought and collaborating with the machine We welcome your feedback
      inkandswitch or helloinkandswitchcom Thank you to everyone who offered
      feedback and guidance throughout the project Peter van Hardenberg James
      Lindenbaum Todd Matthews Kevin Lynagh Geoffrey Litt Scott Jenson Joshua
      Horowitz Patrick Dubroy William Taysom Daniel Krasner Ivan Reese Paul Shen
      Max Schoening
  - text: >-
      Harnessing the Medici Effect for More Profound Web3 Impact CARBON Copy
      Front PageFeatures ReFi Data ProjectsBuildersTokensLandscapeImpact
      DashboardOpportunitiesKnowledge Content NewslettersThe ReFi
      WeeklyLearnWeb3 Fundraising Guide for Nigerian NGOs Resources OnChain
      Grant DirectoryGitcoin Grants CanvasThe Regen AtlasThe Climate Capital
      StackCCN Metrics GardenX List of ReFi ProjectsAboutSubmit News
      OpinionHarnessing the Medici Effect for More Profound Web3 ImpactHow we
      have strayed off the path of originality and how to get back on itBy
      Trinity Morphy May 21st 2024In this piece Trinity Morphy takes inspiration
      from a recent event to look at why originality has been so difficult to
      attain in the Web3 impact space and how a phenomenon called the Medici
      Effect can get us back on trackI had the privilege of attending a Green
      Pill Nigeria Impact Tour event two weeks ago In his talk Izzy the lead at
      Green Pill Nigeria pointed out the troubling trend of imitation in the
      impact space Each cycle he said seems to be filled with recycled project
      ideas It has become customary for Project XYZ to emerge with a seemingly
      novel concept only to be quickly followed by Project ABC essentially a
      copycat with a new bell or whistleThis lack of originality especially in
      Web3 impact is not just a hurdle its a roadblock For instance other
      effective ways exist to utilise impact NFTs apart from sequestering carbon
      or planting new trees Yet we keep seeing new projects based on the same
      idea but with a more exciting descriptionHow can we actively encourage a
      shift from simple recycling to genuine originality The Medici Effect that
      the most innovative ideas happen when you combine concepts from different
      fields offers insight By looking beyond the impact field we can unlock the
      potential for more impactful solutionsWhy we recycle ideas and why
      originality is so importantpictwittercomsRtpkwgMGk— OwockiΞth owocki May
      17 2024Reduced risk and familiarity Humans naturally gravitate towards
      familiarity Replicating a successful project offers security by minimising
      the uncertainty of venturing into something completely new Its like
      following a wellworn path instead of forging one through the
      wildernessAccess to funding Investors and funding institutions are more
      likely to back sectors with a proven track record of projects A copycat
      project can point to the success of the original and argue that it can
      replicate that success with its own twistBandwagon effect When a project
      becomes popular others aim to capitalise on the hype and user base On the
      surface launching a copycat project appears easier as does the marketing
      that goes with itAvoiding mistakes A successful project has already
      learned what works and what does not Emulation enables new projects to
      sidestep those pitfalls and concentrate on innovationOriginality is
      crucial to tackling our social and ecological problems It allows us to see
      things differently question assumptions and uncover new possibilities This
      fresh perspective can lead to new solutions we would not have considered
      otherwise It empowers us to challenge outdated or dysfunctional systems
      and propose solutions that address their shortcomings from the ground up
      Original thinking enables us to recognise the potential in underutilised
      resources and address problems at their roots not simply treat their
      symptomsEasier said than done of course We need to look to the Medici
      Effect for a roadmapThe Medici Effect and Web3 impactThe Medici Effect
      states that the most original and extraordinary ideas occur when you
      combine concepts from different fields disciplines or cultures It was
      discovered by Frans Johansson and shared in a book of the same name A
      majority of the ideas in this section are borrowed from the book and
      modified to fit Web3 impactSome examples of great products that have
      stemmed from the Medici Effect include Silvi reforestation Web3 M3tering
      Protocol renewable energy distribution Web3 Gitcoin fundraising Web3 and
      Toucan Protocol voluntary carbon markets Web3Heres how we can get better
      at using the Medici Effect to our advantageDismantle the barrier between
      Web3 and other fieldsFirst we must be willing to break the associative
      barriers that exist between Web3 and other fields How do we do this
      Through exposure to a diverse set of cultures Culture in this context
      extends beyond geographic boundaries to encompass ethnic class
      professional and organisational differences By immersing ourselves in
      these diverse backgrounds we can unlock a more open and questioning
      mindset and challenge our assumptionsWe also need to embrace broad
      selfeducation Traditional education often compartmentalises knowledge but
      selfdirected exploration across disciplines expands what is possible
      Without it our thinking is limited and we inadvertently stifle creativity
      Selfeducation empowers us to discover unexpected connections and envision
      how concepts in other fields can combine to create groundbreaking Web3
      impact solutionsCombine random conceptsIntersectional ideas are
      groundbreaking because the concepts involved are so different and the
      combinations so unusual that no one would have thought them possible Take
      the inspiration for this article as an example It happened after I came
      across an X account named Cozomo de Medici I was not thinking about Web3
      impact or the Medici Effect That Is how random combinations work You have
      no control over it It just comes Rather than leaving luck to do the work
      we can continuously stimulate our brains to keep producing these random
      combinations How do we achieve thisNurturing our curiosity By obsessing
      over new ideas approaches and perspectives we enhance our brains ability
      to blend random concepts In the foreword of Exploring MycoFi Scott Morris
      pointed to Emmett Jeffs relentless fascination with mushrooms and mycelial
      networks and how it led to the creation of MycoFi a novel cryptoeconomic
      model inspired by the structure of mycelial networksInteracting with
      diverse groups of people Engage a diverse group of people and we will be
      presented with a diverse set of perspectives Interacting with such groups
      can spark unexpected connections For example a conversation between a Web3
      enthusiast and a custodian of tradition might lead to an idea to preserve
      cultural heritage using blockchainThese are what create the conditions for
      the next phaseIgnite and evaluate an explosion of ideasThe strongest
      correlation between quality and quantity of ideas is in fact the number of
      ideas Do you know how many ideas you can get by simply combining concepts
      from environmental science with Web3 concepts A LOT That is why it irks me
      when I see so many projects centering on carbon credit tokenisation when
      there are so many ideas yet to be explored The question here is once these
      ideas start flooding in how do we handle themFirstly capture as many ideas
      as we can Keep track of all the ideas our brains generate and set a target
      to reach before evaluating their feasibility Secondly we need to take our
      time evaluating because our minds will quickly judge the value of an idea
      by comparing it to what is already known to work Its important that we
      evaluate each idea as sincerely as the next whether its gamifying
      traditional classroom learning or eradicating open defecation in
      underserved communitiesComing up with great ideas does not guarantee
      innovation however We must make those ideas happenMake intersectional
      ideas happenThe paradox of innovation at the intersection of fields is
      rooted in the symbiotic relationship between ideas and failures the more
      ideas we explore the higher the rate of failure Far from being a negative
      consequence failures are a vital part of the innovation journey We learn
      and grow through them so that we can do better the next time Not acting
      for fear of failure robs us of this crucial phaseDaring to try ideas opens
      us to a world of invaluable insights that help us refine our approach
      identify weaknesses and ultimately discover the path to a truly
      groundbreaking solution In my interview with Christwin of Switch Electric
      and M3tering Protocol he shared a fascinating story about the early days
      of his solution The first model they tried unexpectedly incentivised the
      consumer to overload the solar infrastructure leading to severe component
      wear and tear and attracting steep maintenance costs This was a
      significant failure but it led to research that brought about the idea of
      building M3tering Protocol on blockchain to track consumption and ensure
      transparencySince failures are inevitable we can allocate specific
      resources for testing each new idea and minimising the potential for
      catastrophic failure This allows for rapid iteration where we can learn
      from each attempt refine our approach and move on to the next experiment
      with valuable insights Do Not forget to document everything along the way
      and share the results within the team This knowledge base becomes a
      valuable resource for future projects because it prevents the same
      mistakes and accelerates future breakthroughs We saw this in the
      transition from Gitcoin 10 and Gitcoin 20We also need to stay motivated
      along the way because the path to ingenuity is rarely linear There will be
      setbacks and failures that make us question our actions Staying motivated
      requires keeping the longterm vision in mind and focusing on the positive
      impact our work can bring to the world We need to celebrate all wins big
      and small whether its successfully completing a pilot project achieving a
      user engagement milestone or receiving positive feedback from a target
      community This reinforces our belief and helps us pushing forward despite
      the inevitable setbacksConclusionThe Web3 impact space has potential to
      change the way we tackle the worlds most pressing challenges Were not
      getting there however if we keep recycling the same old ideas Originality
      is the key if we are to truly unleash Web3 impact We must look beyond our
      own siloes break down barriers and ignite curiosity across a variety of
      fields to spark the kind of breakthrough ideas that will move the
      needleAnd when those ideas do come we must not shy away from failure and
      its valuable lessons Nature exemplifies this beautifully a butterflys
      struggle to escape its chrysalis is anything but pleasant Still this phase
      is necessary for the butterfly to strengthen its muscles expand its wings
      and ultimately fly Embrace the journey celebrate even the smallest wins
      and surround yourself with a supportive network Together we can embrace
      the power of the Medici Effect to create a better future for ourselves and
      the planetThis article represents the opinion of the authors and does not
      necessarily reflect the editorial stance of CARBON CopyCopyright © 2025
      CARBON CopyToken data provided by CoinGeckoFront PageFeaturesLearnReFi
      ProjectsAboutSubmit News
  - text: >-
      5 DeSci projects disrupting scientific research and development — Crypto
      Altruism 0 Skip to Content BLOG CATEGORIES DAOs EDUCATION ENVIRONMENT REFI
      EQUITY INCLUSION FINANCIAL INCLUSION DEFI NFTs PHILANTHROPY SCIENCE DESCI
      SOCIAL IMPACT SUPPLY CHAIN COMMENTARY PODCASTS CRYPTO ALTRUISM PODCAST THE
      WEB3 NONPROFIT IMPACT ON OPTIMISM INFOGRAPHICS RESOURCES BECOME A CRYPTO
      CHARITY DONATING CRYPTO LEVERAGING AI AT YOUR NONPROFIT ABOUT US WHO WE
      ARE TRANSPARENCY AFFILIATE PARTNERSHIPS CONTACT SUPPORT US Open Menu Close
      Menu Open Menu Close Menu BLOG CATEGORIES DAOs EDUCATION ENVIRONMENT REFI
      EQUITY INCLUSION FINANCIAL INCLUSION DEFI NFTs PHILANTHROPY SCIENCE DESCI
      SOCIAL IMPACT SUPPLY CHAIN COMMENTARY PODCASTS CRYPTO ALTRUISM PODCAST THE
      WEB3 NONPROFIT IMPACT ON OPTIMISM INFOGRAPHICS RESOURCES BECOME A CRYPTO
      CHARITY DONATING CRYPTO LEVERAGING AI AT YOUR NONPROFIT ABOUT US WHO WE
      ARE TRANSPARENCY AFFILIATE PARTNERSHIPS CONTACT SUPPORT US BLOG Folder
      CATEGORIES Back DAOs EDUCATION ENVIRONMENT REFI EQUITY INCLUSION FINANCIAL
      INCLUSION DEFI NFTs PHILANTHROPY SCIENCE DESCI SOCIAL IMPACT SUPPLY CHAIN
      COMMENTARY Folder PODCASTS Back CRYPTO ALTRUISM PODCAST THE WEB3 NONPROFIT
      IMPACT ON OPTIMISM INFOGRAPHICS Folder RESOURCES Back BECOME A CRYPTO
      CHARITY DONATING CRYPTO LEVERAGING AI AT YOUR NONPROFIT Folder ABOUT US
      Back WHO WE ARE TRANSPARENCY AFFILIATE PARTNERSHIPS CONTACT SUPPORT US 5
      DeSci projects disrupting scientific research and development Project
      HighlightsScienceDAOs Mar 30 Written By Drew Simon 2021 was the year of
      decentralization and this momentum has only increased into 2022 Not only
      have we seen incredible growth in the decentralized finance DeFi space but
      we have also seen the emergence of social impact DAOs decentralized media
      platforms decentralized VC funds and more recently the emergence of a new
      field – Decentralized Science or DeSci In short “the decentralized science
      DeSci movement aims to harness new technologies such as blockchain and
      ‘Web3’ to address some important research pain points silos and
      bottlenecks” Whereas scientific research has long been viewed as overly
      bureaucratic and disjointed the DeSci movement aims to improve this by
      using blockchain to offer greater transparency and to take on the “profit
      hungry intermediaries” such as scientific journals that have dominated the
      traditional research spaceFor some resources on DeSci I recommend you
      check out the following articlesDeSci an opportunity to decentralize
      scientific research and publicationA Guide to DeSci the Latest Web3
      MovementCall to join the decentralized science movementFor this blog post
      we will be highlighting 5 DeSci projects that are leading the way and
      positively disrupting scientific research and development1 VitaDAOOne of
      the best examples of DeSci in action is VitaDAO a Decentralized Autonomous
      Organization DAO focused on funding longevity research in “an open and
      democratic manner” Specifically they are focused on the decentralization
      of drug development focused on the extension of human life and healthspan
      They fund earlystage research with the goal of turning the research into
      biotech companiesVitaDAO is government by holders of VITA tokens which can
      either be purchased or earned through contributions of work or
      intellectual property With over 4000 members and 9M in funding raised to
      support scientific research VitaDAO has proven that the DeSci movement is
      no laughing matterCheck out some of their featured projects here2
      SCINETThe SCINET platform which is built on blockchain enables retail and
      institutional investors to securely invest in scientific research and
      technology directly In addition to funding promising scientific research
      they also offer a “blockchainpowered” cloud laboratory for researchers a
      rigorous decentralized peer review process and enable researches to
      document their IP on an immutable blockchain3 AntidoteDAOAntidoteDAO is a
      decentralized community focused on funding cancer research and other
      cancer initiatives Their ecosystem includes a governance token and NFT
      collection which both enable individuals to vote on where to allocate
      funds In addition to providing funding to charities supporting cancer
      research and cancer patients a core focus of the DAO is on providing 100K
      seed fund grants to cancer research teams Research projects are first
      reviewed by the DAO’s Medical Advisory team and then put to the community
      for a vote Fun fact we have an upcoming podcast episode with AntidoteDAO
      that when available will be published HERE Crypto Altruism uses Ledger to
      keep its assets safeYou’ve probably heard the phrase “not your keys not
      your coins” By choosing a hard wallet like the Nano S Plus to store your
      crypto you can rest assured that the keys and the crypto are truly
      yoursGet your Ledger Nano S Plus now by clicking HERE or on the image
      below 4 LabDAOLabDAO is an emerging organization which is dedicated to
      operating a communityrun network of wet and dry labs with the goal of
      advancing scientific research and development A wet lab is one focused on
      analysing drugs chemicals and other biological matter whereas a dry lab is
      one focused on applied or computational mathematical analysis LabDAO is a
      relatively new project that is still in its infancy but has a promising
      mission and strong community of support around it 5 MoleculeMolecule is a
      decentralized scientific research funding platform that operates as a
      marketplace for researchers seeking out funding and individuals looking to
      invest in scientific research projects They are “connecting leading
      researchers to funding by turning intellectual property and its
      development into a liquid and easily investable asset”Researchers can list
      their research projects on the Molecule marketplace as a means to engage
      with potential investors and to secure funding for their project Molecule
      currently has over 250 research projects listed on their marketplace over
      4500 DAO community members and 3 “Bio DAOs” with over 10M in funding in
      their network According to Molecule “The future of life science research
      will be driven by open liquid markets for intellectual property powered by
      web3 technology”We cover more amazing DeSci projects in our more recent
      postTen more DeSci projects disrupting scientific research development and
      knowledge sharing Buy me a coffee Send a tip in ETH cryptoaltruismethLike
      what you are reading Consider contributing to Crypto Altruism so we can
      continue putting out great content that shines a light on the good being
      done in the crypto and blockchain community SUPPORT CRYPTO ALTRUISM Please
      note we make use of affiliate marketing to provide readers with referrals
      to high quality and relevant products and services
      DeScidecentralizationscienceblockchainlists Drew Simon Previous Previous
      Crypto Altruism Podcast Episode 39 AntidoteDAO Decentralized funding of
      cancer research and charitable initiatives Next Next Crypto Altruism
      Podcast Episode 38 Using NFTs to empower content creators and help kids
      learn ft Susie Jaramillo CONTENTBLOGPODCASTINFOGRAPHICSCURATED LISTS
      ABOUTABOUTSUPPORT USCONTACTDISCLAIMERPRIVACY POLICY Buy me a coffee
      ETHERC20 cryptoaltruismeth 0xac5C0105914F3afb363699996C9914f193aeDD4A Sign
      up for our monthly newsletter Thank you © Crypto Altruism 2023 FOLLOW
  - text: >-
      A Short History of BiDirectional Links Home The Garden Essays Notes
      Patterns Smidgeons Talks Podcasts Library Antilibrary Now About The Garden
      Essays Notes Patterns Smidgeons Talks Podcasts Library Now About Essays
      evergreen A Short History of BiDirectional Links Seventy years ago we
      dreamed up links that would allow us to create twoway contextual
      conversations Why do not we use them on the web Design Digital Gardening
      The Web Web Development Planted almost 5 years ago Last tended about 4
      years ago Table of Contents Bidirectional links are not new Bidirection
      Linking in Personal Digital Gardens Building Your Own BiDirectionals
      BiDirectional Linking with WebMentions Table of Contents Bidirectional
      links are not new Bidirection Linking in Personal Digital Gardens Building
      Your Own BiDirectionals BiDirectional Linking with WebMentions With the
      recent rise of Roam Research the idea of bidirectional linking is having a
      bit of a moment We are all very used to the monodirectional link the World
      Wide Web is built around They act as oneway pointers we follow in a linear
      sequence While we can link to any site the destination page has not a clue
      we have done so We set up all these singledirection paths trying to signal
      relevance and context only to have the other side completely ignore our
      efforts Our monolinks are trying to establish relationships in vain We are
      starting to look around our monolinked environment and wonder why it is so
      hard to surface relevant contextual relationships Manually interlinking
      content takes an awful lot of human curation and effort Efforts we should
      probably slog off onto our systems Enter the bidirectional link A
      bidirectional link has social awareness it knows about other pages or
      ‘nodes’ that point to it and can make them visible to people This means we
      get a twoway conversation flowing between our web locations Bidirectional
      links are not new The idea of the bidirectional link goes back to 194580ya
      when Vannevar Bush dreamed up the Memex machine Vannevar outlined this
      hypothetical gadget in an essay in The Atlantic called As We May Think He
      wanted a system capable of “associative indexing… whereby any item may be
      caused at will to select immediately and automatically another… so that
      numerous items have been thus joined together to form a trail” This essay
      turned out to be a foundational document for the ideologies that directly
      led to both the internet and the Web Yes those are two entirely separate
      pieces of technology Vannevar was one of the key movers and shakers
      rallying folks to help build the original internet infrastructure He
      corraled folks at MIT the US Department of Defence the National Science
      Foundation and various research labs like the Standford Lincoln Lab Bell
      Labs the RAND Corporation and Xerox PARC to get involved Walter Isaacson
      The Innovators How a Group of Hackers Geniuses and Geeks Created the
      Digital Revolution London Simon Schuster 201510ya Suffice to say the guy
      was driven by a belief that enabling people to connect information and
      share knowledge would expand the scope of human understanding The Memex
      was one idea of how that might manifest in material form Vannevar even
      created a small informative diagram of this deskbound vision Marketing
      chops 101 Vannevar’s evocative description of the Memex is especially
      impressive given that digital computers had only come into existence 5
      years earlier Most were still the domain of large military operations like
      Bletchley Park and were seen as inconveniently large calculators
      Implementing a wildly interactive computational personal knowledge base
      was not much of an option So the idea went into hiberation and did not
      resurface until the idea of personal computing began blooming in the
      sixties and seventies Ted Nelson an unlikely film director and sociologist
      stumbled into a series of computing lectures and began to imagine how
      graphical interfaces might reinvent the way we write and connect ideas He
      took inspiration directly from Vannevar’s essay and in 196560ya when he
      coined the term hypertext to describe his vision for a sprawling network
      of interlinking information Nelson planned to implement these hypertextual
      dreams in his perpetuallyimminent Project Xanadu If you have some time
      this is quite the internet history rabbit hole to run down Ted Nelson is
      on another level The Xanadu project was a hypertext system that imagined
      that every sentence block and page would be part of a vast bidirectionally
      linked network A design mockup of how Project Xanadu might visually
      connect pieces of text across multiple documents You would be able to
      trace information back to its origin the way current web links do But you
      would also be able to see who had referenced remixed and expanded off that
      original The full Pattern Language of Project Xanadu expands far beyond
      just bidirectional links to include features like Transclusions but we
      will not dive into it all here Suffice it to say Xanadu did not pan out
      Instead we got the less fancy but far more real and useable World Wide Web
      that currently does not support bidirectionals on an infrastructure level
      While Sir Tim Berner’s Lee wrote himself a note debating their pros and
      cons back in 199926ya there is an obvious design issue with letting twoway
      connections flow freely around the web If every site that linked to yours
      was visible on your page and you had no control over who could and could
      not link to you it is not hard to imagine the Trollish implications…
      Figuring out how we might filter moderate and set permissions around link
      visibility turned into quite the challenge The design details grew complex
      It became clear implementing the Web with simpler monodirectional links
      was the right thing to do given that its creators wanted universal
      adoption Lots of people are still mad about it Let us not venture too far
      down that historical wormhole The TLDR is technology is hard Until Xanadu
      ships and we are all immersed in the universe of multilinked
      versioncontrolled nodes of remixable microcontent that somehow solves the
      problems of permissions and moderation there are still plenty of ways we
      can resurrect the possibility of bidirectional links on the web
      Bidirection Linking in Personal Digital Gardens Most of the design issues
      with adding bidirectional links to the global web were related to
      moderation and permissions However adding them within the bounds of a
      single website with one author sidesteps that problem There is been a
      flurry of interest around bidirectionals among people involved in the
      Digital Gardening movement Much of this was originally sparked by Andy
      Matuschak’s notes Go have a good browse through them Andys linked notes
      stack on top of one another allowing you to browse to new notes while
      previous notes are still visible There is plenty to admire here It should
      be noted Andy is an experienced developer and interaction designer and
      these notes should not be taken as the standard expectation for the rest
      of us normal plebby internet citizens But the key part of this system that
      creates interlinked context is the “Links to this Note” section at the
      bottom of each post Anytime Andy links to another one of their notes on
      the site it will pop up as a related note at the bottom of the page This
      is the bidirectional dream It gives us a way to navigate through these
      ideas in exploratory mode rather than navigating a hierarchy of categories
      on a main index page Since it is all contained within a singleauthor site
      our SpammishTrollrisk factor is at a comfortable zero This is mildly
      tangential but I love how the topic of bidirectional links makes fully
      visible our “websites are locations” and “websites are containers”
      conceptual metaphors with “inside” and “outside” links Building Your Own
      BiDirectionals That is all very cool but how are you supposed to build
      bidirectionals into your own site Thankfully setting up your own public
      gardening bidirectional Memex does not involve Xanadu One fantastic option
      for nondevelopers is based around a personal wiki system called TiddlyWiki
      AnneLaure Le Cunff wrote up an easytofollow guide to getting your own up
      and running For those of us here for the hypercustomised overengineered
      JavaScript solution that would be me the Gatsbyjs community has a number
      of active gardening enthusiasts building themes and plugins I built mine
      using Aengus McMillin’s gatsbythemebrain Aengus has documented the theme
      well and it is not too challenging to implement as long as you are
      comfortable in JavaScript and React I also curate a list of tools for
      Digital Gardening on this Github repo BiDirectional Linking with
      WebMentions While I argued that Webwide bidirectional links are unlikely
      to happen at a global scale there is a way you can add bidirectionals to
      your personal website that picks up on references anywhere on the web
      WebMentions are a piece of web infrastructure the IndieWeb community has
      done a lot of work to advocate for The W3C gave the specification
      recommendation status in 20178ya The system notifies a URL whenever that
      site is mentioned elsewhere on the web You are then able to show that
      mention and its contents on your site It is essentially an optin
      bidirectional linking system Plenty of folks have written useful guides on
      how to add these to your site Here is one for any static site one for
      Gatsby one for Nextjs There is a whole list of implementation examples on
      the IndieWeb Wiki you can look through 5 Backlinks A Brief History Ethos
      of the Digital Garden A newly revived philosophy for publishing personal
      knowledge on the web Digital Gardening for NonTechnical Folks How to build
      a digital garden without touching code Transclusion and Transcopyright
      Dreams The lost permissioning and copyright system of the Web The Pattern
      Language of Project Xanadu Project Xanadu as a pattern language rather
      than a failed software project A MetaTour of This Site A video tour
      through how I build the old version of this site Mentions around the
      webCarlos Sanmartín Bustosmentioned in En busca de enlaces
      bidireccionalesJune 24 2024Una de las cosas que echo en falta al haber
      pasado el blog a un sitio estático son los links bidireccionales No tanto
      los links que provengan de fuera que hoy en día van a ser mínimos como los
      links internos que permiten explorar el blog por ejemplo para avanzar
      hacia el futurAngsuman Chakrabortymentioned January 07 2023A Short History
      of BiDirectional Links CypherNewsmentioned January 04 2023 Hacker News A
      short history of bidirectional links 2020 hackernews HN Front
      Pagementioned January 04 2023A Short History of BiDirectional Links L C
      newsycombinatorcomitemid342447…Winson Tangmentioned January 04 2023A short
      history of bidirectional links 2020 Hacker Newsmentioned January 04 2023A
      short history of bidirectional links 2020 Patrick Durusau ⏳ White Person
      on White Supremacymentioned May 18 2022codexeditor I Have lost the tweet
      where I saw this mentioned earlier today I like the Web Mentions idea esp
      if we could use XPath to point to less than the page level Thoughts Show 3
      more Want to stay up to date Subscribe via RSS Feed © 2025 Maggie Appleton
      The Garden Essays About Notes Now Patterns Podcasts Talks Smidgeons
      Colophon Library
  - text: >-
      Part 1 Introduction to Responsible Technology by goodbot MediumOpen in
      appSign upSign inWriteSign upSign inPart 1 Introduction to Responsible
      TechnologygoodbotFollow11 min read·Aug 3 2023ListenShareIntroductionIn
      late 2022 the GoodBot team — consisting of a group of sociallyminded
      professionals working in law technology and policy — came together to
      develop a snapshot of Canada’s current Responsible Technology landscape
      This is a space that to date had been heavily defined by voices from the
      United States Our goal is to understand the Canadian ecosystem’s current
      composition capacity and direction particularly how technology impacts
      Canadians what issues are of focus in research and programming and how the
      policy landscape is evolving at the provincial national and international
      levelsTechnology is revolutionizing everything from healthcare to
      education to law to climate science and government but is also associated
      with a wide range of risks and harms making responsible technology
      governance a critical priority for governments nonprofits and marketsNew
      visions clear policy frameworks effective implementation methods and
      multistakeholder oversight bodies are needed to navigate this landscape It
      also requires public interestfocused strategic collaboration that
      includeslongterm research on harmsmore transparency and collaborative
      efforts with technology companies to strengthen safetyeffective mechanisms
      to hold companies accountable when they fail to act in response to
      harminvestment in public interest technology and philosophies that
      prioritize healthy technology ecosystemsunderstanding the leverage points
      and incentive systems that contribute to these outcomes andthe development
      of the critical capacities needed to meet this momentWe know from the last
      two decades that when technology tools and platforms become deeply
      embedded in institutions like media and education before they are
      regulated they become harder to govern This reality has led to escalating
      and lasting harm in a number of areas The pressing question now is how to
      forge a new direction recognizing the immediate need to take actionThis is
      the introduction to a research project on Canada’s Responsible Technology
      landscape GoodBot’s goal with this research is threefoldTo understand the
      current landscape and priorities of Canada’s Responsible Technology
      ecosystem including the who what where and howTo highlight gaps and
      opportunities within this ecosystem that can help Canada develop a more
      robust impactful and collaborative approach and agenda andTo understand
      what role GoodBot can play in advancing Responsible Technology at home and
      around the worldWhat is Responsible TechnologyResponsible Technology acts
      as an umbrella for a range of approaches and terms that all focus on
      different issues or specific intervention points in technology and
      business life cycles It includes concepts such as ethical tech humane tech
      tech stewardship and public interest tech all of which are connected and
      overlap but which also center on different locus of influence These and
      other terms will be explored in Part 1 of our seriesResponsible Technology
      is a relatively new framing that includes a wide range of issues related
      to technology Some issues — like privacy and freedom of expression — are
      longstanding cornerstones of the most established technology nonprofits in
      the country while conversations on Generative AI risks are relatively new
      Yet technology and AI ethics have been around for decades with themes
      prominent among academic labs human rights defenders peacebuilders and in
      other convening spaces What has changed is the scale and pace of
      technology and the amplification of new risks and narratives surrounding
      technology harms These have created a growing awareness of the need for
      safetyfocused design and research at the outset meaningful oversight of
      technology companies and effective accountability mechanisms when
      companies fail to act in the public interestMany technology tools and
      businesses currently fail to meet even a vague understanding of
      Responsible Technology This is especially true for small to medium
      technology companies who focus on survival and leave proactive assessment
      and harm mitigation as an afterthought if they get any attention at all
      Others have social impact strategies that are completely detached from
      their business modelFew companies start with the goal of causing harm but
      unintended consequences can arise due to a lack of intentional
      consideration capacity and unanticipated and conflicting priorities
      Additionally as products scale seemingly harmless matters can lead to real
      harm as user bases grow use cases expand and incentive structures change
      The explosive emergence of generative AI has made it even more clear that
      left unaddressed these structural factors risk widening the gap between
      privatized profit and socialized riskIn spite of these growing concerns
      several Big Tech companies have chosen to lay off large segments of their
      Trust and Safety teams seeing them as cost centers that add undesirable
      operational complexity Even when companies have Trust and Safety teams in
      place they are often pitted against product teams The result is companies
      that increasingly seek to automate these decisions which frequently and
      disproportionately impact minority communities including human rights
      defendersSome companies have begun sharing Transparency Reports which is a
      move in the right direction However there are no agreed standards or
      metrics against which to assess companies’ commitment to social
      sustainability nor is there any external oversight These factors lead to
      the possibility of ‘tech washing’ and cherrypicking data that provide the
      appearance of taking action but in ways that lack substantive effect or
      because new harmMoreover even companies that have the desire to act
      responsibly can lose sight of their original goals when they face demands
      for outsized returns from investors — including venture capitalists and
      private equity investors — which can place them at odds with decisions
      that are in the public interestIn this context a wide range of social
      harms and externalities have arisen — many of which are unintentional —
      and which includeBiased Unaccountable Untransparent AutomationBias in
      untransparent algorithms that discriminate against marginalized
      groupsDisruption of the workforce by generative AI in almost all
      professional sectorsBig Tech DominationBig Tech market domination to
      control value chains through predatory pricing terms and
      acquisitionsNonconsensual selling of personal data to and from thirdparty
      data brokersAddiction Mental HealthThe use of dark patterns to drive
      engagement and addiction in gaming and social mediaThe decline of
      attention spans at a population level in the last 20 yearsThe decline in
      mental health and body image especially among youth leading to self harm
      and deathHarassment Violence ExtremismIncitement to radicalization
      extremism and even genocideTrolling doxing and harassment including
      targeting women trans and BIPOC peopleBad ActorsTargeted and opportunistic
      disinformation and microtargeting to undermine democraciesScams to steal
      millions from people through generative AI or with crypto hacksTrafficking
      of women and girls on the dark web and in mainstream platformsThis is by
      no means a comprehensive list In response Responsible Technology advocates
      have advanced efforts in recent years to understand the wide array of
      externalities impacting different levels of society These initiatives
      variously aim to understand harms and causes increase public awareness and
      engagement incentivize governments to enact new laws and enforce existing
      ones and create solutions that lead to safer and more responsible
      technologyenabled environmentsAdditionally a new wave of government and
      nonprofit investigations and litigation aims to clarify technology
      companies’ responsibilities and identify leverage points to incentivize
      responsibility These efforts have had some successes but are often too
      incremental and underresourced to keep paceThe scale and complexity of
      issues arising from technology are unprecedented Canada needs a clear
      Responsible Technology agenda and sufficient investment to move toward a
      technological future defined by healthy people businesses markets
      societies and democraciesThe Asymmetry Sustainability GapA key barrier
      facing Responsible Technology advocates — including journalists academics
      tech nonprofits technologists tech ethics experts policymakers and
      citizens — in addressing technology harms is an existing and growing set
      of asymmetrical disadvantages when compared to the companies and sectors
      responsible for harmThese disadvantages manifest in many forms including
      limited access to talent data sets algorithms infrastructure information
      internal research and audits knowledge and resources It shows up in access
      to capital restrictions on how capital can be used and comparatively
      robust ethical requirements It is further exacerbated by companies’
      anticompetition tactics to buy out underprice feature bundle and otherwise
      aggressively quash any disruptors who may be able to offer healthier
      alternativesAdditionally asymmetries show up when comparing the inputs and
      outcomes around harm It is — for example — much less resource intensive to
      create campaigns to disseminate disinformation on vaccines — than it is to
      undo the damage caused by this disinformation This reality places a
      societal premium on considering what effective oversight governance and
      accountability of technology look like but also raises the need to balance
      corrective actions with Freedom of Expression FoE norms The global nature
      of many technology platforms means that US norms are effectively imposed
      on Canada and other countries Yet Canada has its own unique and
      wellestablished interpretations of fundamental freedoms that should be
      considered and protected in the face of technological changeCanada’s
      Responsible Technology ecosystem is small and underresourced compared to
      its US counterparts and at this time there are few prominent
      ecosystemlevel organizations that are wellpositioned to guide a
      Responsible Technology community and strategy Yet to effectively influence
      outcomes Responsible Technology advocates need to work together including
      by establishing new governance innovationsA critical factor in the
      Canadian context is that much of the funding to date has focused on
      understanding symptoms and immediate causes rather than underlying
      structural issues and incentive systems at play Some of this is a product
      of new and emerging organizations with limited track records while other
      issues arise from inadequate and restrictive funding and opportunities
      Some organizations only work on technology matters at a project level
      rather than a mission level Such factors are important to understanding
      the limitations on Canada’s capacity and sustainability Our sixth report
      will explore Canada’s nonprofit capacity in more depthAn additional
      barrier is that there are no obvious market solutions to address many of
      the problems that arise from technology For example while technology has
      led to unprecedented online harassment of women people of color and LGBTQ
      communities companies offer limited solutions aimed at limiting
      harassmentThis reality is exacerbated by the fact that markets recently
      rewarded technology companies for cuts that significantly depleted their
      Trust Safety teams even when it occurred at a time of high salience around
      the risks posed by technology platformsLack of coordination is also a
      factor among tech companies For example some companies argue that if they
      do not employ harmful tactics such as polarizing content or algorithms to
      attract attention they will lose out to rivals who will These realities
      point to an increasing need for sectors to collaborate toward reducing
      harm and promoting public interestIndeed in areas where companies have
      invested more material resources — including to address extremism and
      protect children — even Big Tech lacks the bandwidth to address the
      complex range of issues on its own making collaboration essential Within
      the tech sector new multistakeholder initiatives such as Global Internet
      Forum to Counter Terrorism and the Tech Coalition have been launched in an
      effort by technology platforms — and include human rights advocates
      governments and researchers — who work collectively to reduce extremism
      and child sexual abuse material online respectively Such issues are even
      more challenging for smaller platforms and startups that lack internal
      resources to respond to emerging and unanticipated issuesUltimately while
      technology companies contribute many benefits to society they also
      contribute significant problems that neither society the markets nor they
      are presently able to solve They often lack incentives to prevent and
      address problems up front and even when they want to do the right thing
      investor incentives can derail their decisions There are limited market
      incentives to address challenges created by businesses and market
      competition Moreover our governance institutions operate in ways that are
      incompatible with the fastmoving pace of technology which presently tend
      to focus on harm to individuals instead of on harm to society These
      factors together mean that we are in a situation that is by definition
      unsustainable We need collective action to address these risks including
      through responsible resources capacities frameworks budgets and
      policiesCanada’s Policy LandscapeWhile several technologyfocused bills are
      in development Canada’s current policy landscape is far behind the
      advances of the last 20 years This is also true of many other countries
      that currently lack the policies capacities institutions and enforcement
      mechanisms needed to govern a rapidly evolving technology environment
      Furthermore research shows that even where effective policies are in place
      underresourced enforcement mechanisms such as antitrust hamper the ability
      of governments to enforce existing laws In this context the role of civil
      society organizations is particularly importantWhile the policy landscape
      is expected to evolve rapidly we do not know what effect these policies
      will have or how coherent they will be across jurisdictions and issues
      Canada has introduced many bills for consideration but few have passed and
      all leave much to be desired In the intervening time legal firms and the
      Privacy Commissioner have advanced litigation — with a particular focus on
      privacy and antitrust — to hold technology companies accountableCivil
      society organizations often lack consensus on addressing key issues For
      instance Online Harms advocates support stricter content moderation to
      tackle harassment extremism and child exploitation In Canada some lean
      towards Freedom of Expression while others recognize the need to address
      such harms but fear that wellintentioned policies could inadvertently
      suppress the voices of vulnerable communities Balancing these competing
      rights is complex Responding to emerging issues and harms that affect us
      all requires better public engagement This includes open and inclusive
      dialogue transparent consultation processes and effective accountability
      mechanisms to navigate these complexities and to help uphold and balance a
      range of fundamental Canadian rightsIndeed there are no ‘right’ answers
      but rather ‘different tradeoffs’ Moreover there are ways to escape such
      polarity traps which can easily become politicized resulting in deadlock
      Even among nonprofit organizations such as the ones that are seemingly at
      odds on online harms for example both largely agree that passing
      comprehensive privacy frameworks would mark an important victory and
      achievementYet even if we achieve effective regulation and enforcement
      addressing entrenched asymmetries — especially those caused by Big Tech —
      requires a collective agenda and roadmap What is clear is that our current
      institutions lack the capacity and resilience needed to address the
      challenge we faceAs Canadians we have an opportunity to draw upon the
      values that make us strong as we reimagine our relationship with
      technology Ideas such as Indigenous approaches to data sovereignty
      collaboration and multiculturalism have much to teach us about how to
      navigate these complex issues This moment presents an opportunity to
      rethink and localize the who how and why of technology governance This is
      both a daunting and an exciting challengeOpenSourcing GoodBot’s
      ResearchFor GoodBot our first step is to practice our open principles by
      sharing what we have learned We hope that this research can lay the
      groundwork for building a Canadian coalition to support its nascent and
      necessary Responsible Technology movementGetting to impact requires
      understanding Canada’s existing capacity understanding the systemic issues
      at play exploring moral and policy considerations surfacing current and
      emerging asymmetries of power and exploring how AI is upending companies
      and industries It also requires a collective strategy and targeted action
      focused on moving toward responsibilityThese documents are intended to act
      as a primer for anyone seeking to make an impact in addressing critical
      priorities facing Canada and the world While our early research may
      initially be of more value to nonprofits and academics and should be
      considered a WorkinProgress we aspire to a future where solutionsoriented
      multistakeholder collaboration is a new norm Our research will be broken
      into two partsPart 1 is a Canadian Responsible Technology Landscape that
      explores common terms civil society stakeholders current and emerging
      policies and litigation and how asymmetries of power manifestPart 2
      reviews the results of a survey conducted key observations on the current
      and emerging landscape and critical reflections on how to strengthen
      interdisciplinary collaboration among nonprofit organizations academia the
      tech sector and governmentIn an ideal world this work will lead to
      highlevel consultations and strategies developed in collaboration with
      other ecosystem organizations motivated to move this conversation
      forwardCanada as a Global Leader in Responsible TechDespite and perhaps
      because of the wide array of challenges Canada has an opportunity to
      become a global leader in developing deploying and governing technology in
      socially sustainable ways Getting there requires an urgent focus on
      strengthening national capabilities by investing in strategic and
      systemsfocused multistakeholder mechanismsIndeed organized effectively
      Canadian civil society represents critical and untapped assets to help
      meet this moment There is also a need to strengthen citizen education
      advance responsible policy and oversight create technical solutions to
      advance the public interest introduce responsible technology certifiers
      and respond to systemic factors that lead to harmful outcomes Canadians
      can no longer afford to wait The time to engage is nowVersion 10 July 2023
      Written by Renee BlackFollowWritten by goodbot3 Followers·1
      FollowingFollowNo responses
      yetHelpStatusAboutCareersPressBlogPrivacyRulesTermsText to speech
metrics:
  - accuracy
pipeline_tag: text-classification
library_name: setfit
inference: false
base_model: sentence-transformers/paraphrase-mpnet-base-v2
model-index:
  - name: SetFit with sentence-transformers/paraphrase-mpnet-base-v2
    results:
      - task:
          type: text-classification
          name: Text Classification
        dataset:
          name: Unknown
          type: unknown
          split: test
        metrics:
          - type: accuracy
            value: 0.5517241379310345
            name: Accuracy

SetFit with sentence-transformers/paraphrase-mpnet-base-v2

This is a SetFit model that can be used for Text Classification. This SetFit model uses sentence-transformers/paraphrase-mpnet-base-v2 as the Sentence Transformer embedding model. A MultiOutputClassifier instance is used for classification.

The model has been trained using an efficient few-shot learning technique that involves:

  1. Fine-tuning a Sentence Transformer with contrastive learning.
  2. Training a classification head with features from the fine-tuned Sentence Transformer.

Model Details

Model Description

Model Sources

Evaluation

Metrics

Label Accuracy
all 0.5517

Uses

Direct Use for Inference

First install the SetFit library:

pip install setfit

Then you can load this model and run inference.

from setfit import SetFitModel

# Download from the 🤗 Hub
model = SetFitModel.from_pretrained("praisethefool/human_tech-fields-multilabelclassifier")
# Run inference
preds = model("5 DeSci projects disrupting scientific research and development — Crypto Altruism 0 Skip to Content BLOG CATEGORIES DAOs EDUCATION ENVIRONMENT REFI EQUITY INCLUSION FINANCIAL INCLUSION DEFI NFTs PHILANTHROPY SCIENCE DESCI SOCIAL IMPACT SUPPLY CHAIN COMMENTARY PODCASTS CRYPTO ALTRUISM PODCAST THE WEB3 NONPROFIT IMPACT ON OPTIMISM INFOGRAPHICS RESOURCES BECOME A CRYPTO CHARITY DONATING CRYPTO LEVERAGING AI AT YOUR NONPROFIT ABOUT US WHO WE ARE TRANSPARENCY AFFILIATE PARTNERSHIPS CONTACT SUPPORT US Open Menu Close Menu Open Menu Close Menu BLOG CATEGORIES DAOs EDUCATION ENVIRONMENT REFI EQUITY INCLUSION FINANCIAL INCLUSION DEFI NFTs PHILANTHROPY SCIENCE DESCI SOCIAL IMPACT SUPPLY CHAIN COMMENTARY PODCASTS CRYPTO ALTRUISM PODCAST THE WEB3 NONPROFIT IMPACT ON OPTIMISM INFOGRAPHICS RESOURCES BECOME A CRYPTO CHARITY DONATING CRYPTO LEVERAGING AI AT YOUR NONPROFIT ABOUT US WHO WE ARE TRANSPARENCY AFFILIATE PARTNERSHIPS CONTACT SUPPORT US BLOG Folder CATEGORIES Back DAOs EDUCATION ENVIRONMENT REFI EQUITY INCLUSION FINANCIAL INCLUSION DEFI NFTs PHILANTHROPY SCIENCE DESCI SOCIAL IMPACT SUPPLY CHAIN COMMENTARY Folder PODCASTS Back CRYPTO ALTRUISM PODCAST THE WEB3 NONPROFIT IMPACT ON OPTIMISM INFOGRAPHICS Folder RESOURCES Back BECOME A CRYPTO CHARITY DONATING CRYPTO LEVERAGING AI AT YOUR NONPROFIT Folder ABOUT US Back WHO WE ARE TRANSPARENCY AFFILIATE PARTNERSHIPS CONTACT SUPPORT US 5 DeSci projects disrupting scientific research and development Project HighlightsScienceDAOs Mar 30 Written By Drew Simon 2021 was the year of decentralization and this momentum has only increased into 2022 Not only have we seen incredible growth in the decentralized finance DeFi space but we have also seen the emergence of social impact DAOs decentralized media platforms decentralized VC funds and more recently the emergence of a new field – Decentralized Science or DeSci In short “the decentralized science DeSci movement aims to harness new technologies such as blockchain and ‘Web3’ to address some important research pain points silos and bottlenecks” Whereas scientific research has long been viewed as overly bureaucratic and disjointed the DeSci movement aims to improve this by using blockchain to offer greater transparency and to take on the “profit hungry intermediaries” such as scientific journals that have dominated the traditional research spaceFor some resources on DeSci I recommend you check out the following articlesDeSci an opportunity to decentralize scientific research and publicationA Guide to DeSci the Latest Web3 MovementCall to join the decentralized science movementFor this blog post we will be highlighting 5 DeSci projects that are leading the way and positively disrupting scientific research and development1 VitaDAOOne of the best examples of DeSci in action is VitaDAO a Decentralized Autonomous Organization DAO focused on funding longevity research in “an open and democratic manner” Specifically they are focused on the decentralization of drug development focused on the extension of human life and healthspan They fund earlystage research with the goal of turning the research into biotech companiesVitaDAO is government by holders of VITA tokens which can either be purchased or earned through contributions of work or intellectual property With over 4000 members and 9M in funding raised to support scientific research VitaDAO has proven that the DeSci movement is no laughing matterCheck out some of their featured projects here2 SCINETThe SCINET platform which is built on blockchain enables retail and institutional investors to securely invest in scientific research and technology directly In addition to funding promising scientific research they also offer a “blockchainpowered” cloud laboratory for researchers a rigorous decentralized peer review process and enable researches to document their IP on an immutable blockchain3 AntidoteDAOAntidoteDAO is a decentralized community focused on funding cancer research and other cancer initiatives Their ecosystem includes a governance token and NFT collection which both enable individuals to vote on where to allocate funds In addition to providing funding to charities supporting cancer research and cancer patients a core focus of the DAO is on providing 100K seed fund grants to cancer research teams Research projects are first reviewed by the DAO’s Medical Advisory team and then put to the community for a vote Fun fact we have an upcoming podcast episode with AntidoteDAO that when available will be published HERE Crypto Altruism uses Ledger to keep its assets safeYou’ve probably heard the phrase “not your keys not your coins” By choosing a hard wallet like the Nano S Plus to store your crypto you can rest assured that the keys and the crypto are truly yoursGet your Ledger Nano S Plus now by clicking HERE or on the image below 4 LabDAOLabDAO is an emerging organization which is dedicated to operating a communityrun network of wet and dry labs with the goal of advancing scientific research and development A wet lab is one focused on analysing drugs chemicals and other biological matter whereas a dry lab is one focused on applied or computational mathematical analysis LabDAO is a relatively new project that is still in its infancy but has a promising mission and strong community of support around it 5 MoleculeMolecule is a decentralized scientific research funding platform that operates as a marketplace for researchers seeking out funding and individuals looking to invest in scientific research projects They are “connecting leading researchers to funding by turning intellectual property and its development into a liquid and easily investable asset”Researchers can list their research projects on the Molecule marketplace as a means to engage with potential investors and to secure funding for their project Molecule currently has over 250 research projects listed on their marketplace over 4500 DAO community members and 3 “Bio DAOs” with over 10M in funding in their network According to Molecule “The future of life science research will be driven by open liquid markets for intellectual property powered by web3 technology”We cover more amazing DeSci projects in our more recent postTen more DeSci projects disrupting scientific research development and knowledge sharing Buy me a coffee Send a tip in ETH cryptoaltruismethLike what you are reading Consider contributing to Crypto Altruism so we can continue putting out great content that shines a light on the good being done in the crypto and blockchain community SUPPORT CRYPTO ALTRUISM Please note we make use of affiliate marketing to provide readers with referrals to high quality and relevant products and services DeScidecentralizationscienceblockchainlists Drew Simon Previous Previous Crypto Altruism Podcast Episode 39 AntidoteDAO Decentralized funding of cancer research and charitable initiatives Next Next Crypto Altruism Podcast Episode 38 Using NFTs to empower content creators and help kids learn ft Susie Jaramillo CONTENTBLOGPODCASTINFOGRAPHICSCURATED LISTS ABOUTABOUTSUPPORT USCONTACTDISCLAIMERPRIVACY POLICY Buy me a coffee ETHERC20 cryptoaltruismeth 0xac5C0105914F3afb363699996C9914f193aeDD4A Sign up for our monthly newsletter Thank you © Crypto Altruism 2023 FOLLOW")

Training Details

Training Set Metrics

Training set Min Median Max
Word count 20 2568.9241 13352

Training Hyperparameters

  • batch_size: (8, 8)
  • num_epochs: (1, 1)
  • max_steps: -1
  • sampling_strategy: oversampling
  • body_learning_rate: (2e-05, 1e-05)
  • head_learning_rate: 0.01
  • loss: CosineSimilarityLoss
  • distance_metric: cosine_distance
  • margin: 0.25
  • end_to_end: False
  • use_amp: True
  • warmup_proportion: 0.1
  • l2_weight: 0.01
  • seed: 42
  • evaluation_strategy: steps
  • eval_max_steps: -1
  • load_best_model_at_end: True

Training Results

Epoch Step Training Loss Validation Loss
0.0017 1 0.2236 -
0.0694 40 - 0.1379
0.0868 50 0.1722 -
0.1389 80 - 0.1440
0.1736 100 0.0536 -
0.2083 120 - 0.1412
0.2604 150 0.0293 -
0.2778 160 - 0.1343
0.3472 200 0.0234 0.1406
0.4167 240 - 0.1266
0.4340 250 0.0176 -
0.4861 280 - 0.1118
0.5208 300 0.0193 -
0.5556 320 - 0.1095
0.6076 350 0.0162 -
0.625 360 - 0.0926
0.6944 400 0.0223 0.0995
0.7639 440 - 0.0923
0.7812 450 0.018 -
0.8333 480 - 0.0814
0.8681 500 0.0045 -
0.9028 520 - 0.0801
0.9549 550 0.0074 -
0.9722 560 - 0.0794

Framework Versions

  • Python: 3.11.12
  • SetFit: 1.1.2
  • Sentence Transformers: 3.4.1
  • Transformers: 4.51.3
  • PyTorch: 2.6.0+cu124
  • Datasets: 3.5.1
  • Tokenizers: 0.21.1

Citation

BibTeX

@article{https://doi.org/10.48550/arxiv.2209.11055,
    doi = {10.48550/ARXIV.2209.11055},
    url = {https://arxiv.org/abs/2209.11055},
    author = {Tunstall, Lewis and Reimers, Nils and Jo, Unso Eun Seo and Bates, Luke and Korat, Daniel and Wasserblat, Moshe and Pereg, Oren},
    keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences},
    title = {Efficient Few-Shot Learning Without Prompts},
    publisher = {arXiv},
    year = {2022},
    copyright = {Creative Commons Attribution 4.0 International}
}