text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Understanding Batch Normalization
Batch normalization is a technique which has been successfully applied to neural networks ever since it was introduced in 2015. Empirically, it decreases training time and helps maintain the stability of deep neural networks. For that reason practitioners have adopted the technique ... | https://www.lesswrong.com/posts/T7tHHt9Hv7ntFRHTL/understanding-batch-normalization |
# Why Subagents?
The justification for modelling real-world systems as “agents” - i.e. choosing actions to maximize some utility function - usually rests on [various coherence theorems](https://www.lesswrong.com/posts/RQpNHSiWaXTvDxt6R/coherent-decisions-imply-consistent-utilities). They say things like “either the sy... | https://www.lesswrong.com/posts/3xF66BNSC5caZuKyC/why-subagents |
# [Site Update] Weekly/Monthly/Yearly on All Posts
Last week, our friends over at the EA Forum coded a new feature for the /allPosts page – in addition to the daily view, you can now view posts by weekly, monthly and yearly. (Thanks JP!)
This is most exciting when you also set the sorting to "top karma", making it an... | https://www.lesswrong.com/posts/rtrQTCBbKTHiQhRJP/site-update-weekly-monthly-yearly-on-all-posts |
# Open problems in human rationality: guesses
A couple months back Raemon wrote this [excellent question](https://www.lesswrong.com/posts/gX8fcAwk3HGkFyJgk/what-are-the-open-problems-in-human-rationality) to which Scott Alexander shared his [ongoing list](https://www.lesswrong.com/posts/gX8fcAwk3HGkFyJgk/what-are-the-... | https://www.lesswrong.com/posts/sJjmgQnTLE2j3gKTt/open-problems-in-human-rationality-guesses |
# Permissions in Governance
**Compliance Costs**
The burden of a rule can be separated into (at least) two components.
First, there’s the direct opportunity cost of not being allowed to do the things the rule forbids. (We can include here the penalties for violating the rule.)
Second, there’s the “cost of complianc... | https://www.lesswrong.com/posts/jb9zbNr6mkJRiMcjk/permissions-in-governance |
# Rethinking Batch Normalization
Yesterday we saw a glimpse into the inner workings of batch normalization, a popular technique in the field of deep learning. Given that the effectiveness of batch normalization has been demonstrated beyond any reasonable doubt, it may come as a surprise that researchers don't really k... | https://www.lesswrong.com/posts/aLhuuNiLCrDCF5QTo/rethinking-batch-normalization |
# Very different, very adequate outcomes
Let $U_p$ be the utility function that - somehow - expresses your preferences[^pref]. Let $U_h$ be the utility function expresses your hedonistic pleasure.
[^pref]: This being a somewhat naive form of preference utilitarianism, along the lines of "if the human choose it, then ... | https://www.lesswrong.com/posts/2HWxzYqo46dve5Ln7/very-different-very-adequate-outcomes |
# Power Buys You Distance From The Crime
Introduction
------------
Taxes are typically meant to be proportional to money (or negative externalities, but that's not what I'm focusing on). But one thing money buys you ... | https://www.lesswrong.com/posts/9fB4gvoooNYa4t56S/power-buys-you-distance-from-the-crime |
# Practical consequences of impossibility of value learning
There is a [No Free Lunch result](https://arxiv.org/abs/1712.05812) in value-learning. Essentially, you can't learn the preferences of an agent from its behaviour unless you make assumptions about its rationality, and you can't learn its rationality unless yo... | https://www.lesswrong.com/posts/cnjWN4mzmWzggRnCJ/practical-consequences-of-impossibility-of-value-learning |
# Open & Welcome Thread - August 2019
* If it’s worth saying, but not worth its own post, here's a place to put it.
* You can also make a [shortform post](http://www.lesslong.com).
* And, if you are new to LessWrong, here's the place to introduce yourself.
* Personal stories, anecdotes, or just general comm... | https://www.lesswrong.com/posts/HPxTwyrgzCRWWoCPh/open-and-welcome-thread-august-2019 |
# Proposed algorithm to fight anchoring bias
[Anchoring](https://en.wikipedia.org/wiki/Anchoring) is a classic cognitive bias which has been [discussed on Less Wrong before](https://www.lesswrong.com/posts/bMkCEZoBNhgRBtzoj/anchoring-and-adjustment). Anchoring seems very [difficult to avoid](https://en.wikipedia.org/... | https://www.lesswrong.com/posts/BtPcgHZspX32BskTY/proposed-algorithm-to-fight-anchoring-bias |
# Alleviating Bipolar with meditation
Original post: [Alleviating Bipolar with meditation](http://bearlamp.com.au/alleviating-bipolar-with-meditation/)
I was asked on [the slack](https://wiki.lesswrong.com/wiki/Less_Wrong_Slack), about bipolar and what might help from a meditation standpoint. I have my own experie... | https://www.lesswrong.com/posts/qgLyETFo683TqMYNA/alleviating-bipolar-with-meditation |
# Zeno walks into a bar
Zeno walks into a bar.
"I have a problem," he said.
"What is it?" said the bartender.
"Well, it has to do with the movement of physical bodies," said Zeno.
"Talk to my friend Max," said the bartender. He gestured toward a German man wearing round spectacles.
"Sir," said... | https://www.lesswrong.com/posts/m3KS535iRffYwhS3p/zeno-walks-into-a-bar |
# Cephaloponderings
[Cross-posted from Putanumonit](https://putanumonit.com/2019/07/26/cephaloponderings/).
* * *
Hello all. This is Jacob’s wife, and [while Jacob is off chilling in the fjords](https://putanumonit.com/2019/07/20/summer-programming-notice-2019/), I’m staying home and have volunteered to write a gues... | https://www.lesswrong.com/posts/4qzoXiJdSrLsFNtWW/cephaloponderings |
# Inversion of theorems into definitions when generalizing
This post describes a pattern of abstraction that is common in mathematics, which I haven't seen described in explicit terms elsewhere. I would appreciate pointers to any existing discussions. Also, I would appreciate more examples of this phenomenon, as well ... | https://www.lesswrong.com/posts/rs2focaRymwvkW2jS/inversion-of-theorems-into-definitions-when-generalizing |
# Can we really prevent all warming for less than 10B$ with the mostly side-effect free geoengineering technique of Marine Cloud Brightening?
If we're to believe the [Philosophical Transactions of the Royal Society](https://royalsocietypublishing.org/doi/full/10.1098/rsta.2012.0086), or the [Copenhagen Consensus Cente... | https://www.lesswrong.com/posts/XRTiojqqJ3wrFFZAf/can-we-really-prevent-all-warming-for-less-than-10busd-with |
# Do you do weekly or daily reviews? What are they like?
* What system(s) do you use to keep yourself organized and working toward your goals? I'm interested in technologies, but more interested in what ontology you use to organize your tasks, events, goals, and plans.
* Did you build your system slowly over time, or ... | https://www.lesswrong.com/posts/duXd7HMXcPJLdANRP/do-you-do-weekly-or-daily-reviews-what-are-they-like |
# [AN #61] AI policy and governance, from two people in the field
Find all Alignment Newsletter resources [here](http://rohinshah.com/alignment-newsletter/). In particular, you can [sign up](http://eepurl.com/dqMSZj), or look through this [spreadsheet](https://docs.google.com/spreadsheets/d/1PwWbWZ6FPqAgZWOoOcXM8N_tUC... | https://www.lesswrong.com/posts/ZtXMM78zTqBe3eduo/an-61-ai-policy-and-governance-from-two-people-in-the-field |
# My recommendations for gratitude exercises
Gratitude has become an increasingly important part of my life. It has also been one of my greatest sources of improvement of well-being and one of the biggest factors in lifting me out of depression. How does this work? The short answer is that I keep a gratitude journ... | https://www.lesswrong.com/posts/uhuLCNWyDyzvMQD8C/my-recommendations-for-gratitude-exercises |
# How to navigate through contradictory (health/fitness) advice?
I will start with a brief story, but the question can be generalized.
Last year, I decided to do something for my body. I joined and regularly went to K. Training (abbreviated name), a large gym chain in german-speaking countries. The claimed philosophy... | https://www.lesswrong.com/posts/fpdP7maTzcLLXBCjb/how-to-navigate-through-contradictory-health-fitness-advice |
# Preferences as an (instinctive) stance
User Veedrac [recently commented](https://www.lesswrong.com/posts/cnjWN4mzmWzggRnCJ/practical-consequences-of-impossibility-of-value-learning#kPMSLqAJn7b5Kptq6):
>You have shown that simplicity cannot distinguish $(p, R)$ from $(-p, -R)$, but you have not shown that simplicity... | https://www.lesswrong.com/posts/f5p7AiDkpkqCyBnBL/preferences-as-an-instinctive-stance |
# A Survey of Early Impact Measures
In the context of AI alignment an impact penalty is one way of avoiding large negative side effects from misalignment. The idea is that rather than specifying negative impacts, we can try to avoid catastrophes by avoiding large side effects altogether.
Impact measures are ways to m... | https://www.lesswrong.com/posts/TPy4RJvzogqqupDKk/a-survey-of-early-impact-measures |
# What are the best resources for examining the evidence for anthropogenic climate change?
A while back I was researching the evidence for evolution. It's not that I didn't initially believe in evolution or understand natural selection, but it's just that I didn't really understand the full breadth of the evidence and... | https://www.lesswrong.com/posts/KPjet885EeDFEYhj4/what-are-the-best-resources-for-examining-the-evidence-for |
# Status 451 on Diagnosis: Russell Aphasia
Samuel the Fifth at _Status 451_ considers the implications of [Russell conjugation](https://www.edge.org/response-detail/27181), or [emotive conjugation](https://en.wikipedia.org/wiki/Emotive_conjugation) (example: she lied; you distorted the truth; I [honestly reported unco... | https://www.lesswrong.com/posts/etAZs3QG8LZwm4bTB/status-451-on-diagnosis-russell-aphasia |
# Why is the nitrogen cycle so under-emphasized compared to climate change
While listing various environmental issues of how [planetary boundaries](https://www.stockholmresilience.org/research/planetary-boundaries/planetary-boundaries/about-the-research/the-nine-planetary-boundaries.html) are exceeded the Stockholm Re... | https://www.lesswrong.com/posts/wz6Bf2JBNjvazmb7M/why-is-the-nitrogen-cycle-so-under-emphasized-compared-to |
# Trauma, Meditation, and a Cool Scar
\[Trigger Warning: I’ll be discussing a physical injury, recovery, and panic attacks in detail. The first three pictures linked are gory. Again, they are linked, not directly shown\]
Trauma
======
One year ago today, I was in an accident with an industrial drone. It was spinning... | https://www.lesswrong.com/posts/doiMq8aH2yiZaCJsT/trauma-meditation-and-a-cool-scar |
# Weak foundation of determinism
analysis
Determinism is the belief that every action in time is born by the previous one. \[[1](https://en.wikipedia.org/wiki/Determinism)\] In deterministic terms you cannot have an event _**E**_ if you didn't have an event _**D**_ before, which in turn is the result of a cause _**C... | https://www.lesswrong.com/posts/Fm2TMHpkGnPCBxwmv/weak-foundation-of-determinism-analysis |
# New paper: Corrigibility with Utility Preservation
I am pleased to announce the availability of a long-format paper with new results on AGI safety: **Corrigibility with Utility Preservation**.
You can get the paper at [https://arxiv.org/abs/1908.01695](https://arxiv.org/abs/1908.01695) , and in the related soft... | https://www.lesswrong.com/posts/3uHgw2uW6BtR74yhQ/new-paper-corrigibility-with-utility-preservation |
# Subagents, neural Turing machines, thought selection, and blindspots
In my [summary of Consciousness and the Brain](https://www.lesswrong.com/posts/x4n4jcoDP7xh5LWLq/book-summary-consciousness-and-the-brain) (Dehaene, 2014), I briefly mentioned that one of the functions of consciousness is to [carry out artificial s... | https://www.lesswrong.com/posts/7zQPYQB5EeaqLrhBh/subagents-neural-turing-machines-thought-selection-and |
# Project Proposal: Considerations for trading off capabilities and safety impacts of AI research
There seems to be some amount of consensus that people working on AI safety (at least within fairly mainstream ML / AI paradigms) shouldn't worry much about the effects of their projects on AI capabilities. New researcher... | https://www.lesswrong.com/posts/y5fYPAyKjWePCsq3Y/project-proposal-considerations-for-trading-off-capabilities |
# Understanding Recent Impact Measures
In the first five years after Stuart Armstrong posted his first [research suggestions](https://www.lesswrong.com/posts/8Nwg7kqAfCM46tuHq/the-mathematics-of-reduced-impact-help-needed) for impact measures, very little published work expanded on the idea. The [last post](https://ww... | https://www.lesswrong.com/posts/pf48kg9xCxJAcHmQc/understanding-recent-impact-measures |
# Emotions are not beliefs
This lesson took me a long time to learn. Consider the following questions:
Should I be confident? Do you regret that decision? Are you happy about what happened?
In each case, there are two things that are being asked:
* Do I have a good chance of success? AND Should I feel confident?
... | https://www.lesswrong.com/posts/8yp8ixrb6zj5p9bMK/emotions-are-not-beliefs |
# Self-Supervised Learning and AGI Safety
(*Update later on: I'm no longer personally excited about this direction, for reasons mentioned in [section 7.2 here](https://www.lesswrong.com/posts/Gfw7JMdKirxeSPiAk/solving-the-whole-agi-control-problem-version-0-0001).*)
**Abstract:** We should seriously consider the poss... | https://www.lesswrong.com/posts/EMZeJ7vpfeF4GrWwm/self-supervised-learning-and-agi-safety |
# Help forecast study replication in this social science prediction market
_This is a linkpost for: [https://www.replicationmarkets.com/](https://www.replicationmarkets.com/)_
From their sign-up page:
> Help the Replication Markets team spot reliable science, and win $$!
> From mid-2019 to mid-2020, we will run pre... | https://www.lesswrong.com/posts/XzetppcF8BNoDqFBs/help-forecast-study-replication-in-this-social-science |
# In defense of Oracle ("Tool") AI research
(*Update 2022:* Enjoy the post, but note that it’s old, has some errors, and is certainly not reflective of my current thinking. –Steve)
*Low confidence; offering this up for discussion*
An Oracle AI is an AI that only answers questions, and doesn't take any other actions.... | https://www.lesswrong.com/posts/AKtn6reGFm5NBCgnd/in-defense-of-oracle-tool-ai-research |
# Four Ways An Impact Measure Could Help Alignment
Impact penalties are designed to help prevent an artificial intelligence from taking actions which are catastrophic.
Despite the apparent simplicity of this approach, there are in fact a plurality of different frameworks under which impact measures could prove helpfu... | https://www.lesswrong.com/posts/wJK944YqvFwjdbqCP/four-ways-an-impact-measure-could-help-alignment |
# Toy model piece #2: Combining short and long range partial preferences
I'm working towards a toy model that will illustrate all the steps in the [research agenda](https://www.lesswrong.com/posts/CSEdLLEkap2pubjof/research-agenda-v0-9-synthesising-a-human-s-preferences-into). It will start with some algorithmic stand... | https://www.lesswrong.com/posts/gzwM9edW4fXPwJuAW/toy-model-piece-2-combining-short-and-long-range-partial |
# Verification and Transparency
_Epistemic status: I’ve thought about this topic in general for a while, and recently spent half an hour thinking about it in a somewhat focussed way._
Verification and transparency are two kinds of things you can do to or with a software system. Verification is where you use a program... | https://www.lesswrong.com/posts/n3YRDJYCnQcDAw29G/verification-and-transparency |
# Calibrating With Cards
In this post, I'll try to bring together two things I enjoy: rationality and magic. Like [Hazard](https://www.lesswrong.com/posts/ctBCPaD3yYimEHXPu/magic-is-dead-give-me-attention), I've also practiced close-up magic for a [good amount of time now](https://www.youtube.com/watch?v=JfmIY5RG8gQ).... | https://www.lesswrong.com/posts/Ktw9L67Nwq8o2kecP/calibrating-with-cards |
# Which of these five AI alignment research projects ideas are no good?
I'll post five AI alignment research project ideas as comments. It would be
great if you could approval-vote on them by using upvotes. Ie. when you think
the project idea isn't good, you leave the comment as is; otherwise you give it
a single upvo... | https://www.lesswrong.com/posts/m84D5GZMMERvH53q3/which-of-these-five-ai-alignment-research-projects-ideas-are |
# Why do humans not have built-in neural i/o channels?
Communication between organisms of the same species is often beneficial, for a variety of reasons: sharing information, signalling, bonding, etc. Yet currently the most advanced form of communication to have evolved, human language, is still very low bandwidth com... | https://www.lesswrong.com/posts/L9u3kyE6mSsCCKnc3/why-do-humans-not-have-built-in-neural-i-o-channels |
# AI Forecasting Dictionary (Forecasting infrastructure, part 1)
_This post introduces the [AI Forecasting Dictionary](https://parallel-forecast.github.io/AI-dict/), an open-source set of standards and conventions for precisely interpreting AI and auxiliary terms. It is the first part in a series of blog posts which m... | https://www.lesswrong.com/posts/8y7DcSF4eAkXoru4u/ai-forecasting-dictionary-forecasting-infrastructure-part-1-2 |
# Why Gradients Vanish and Explode
_Epistemic status: Confused, but trying to explain a concept that I previously thought I understood. I suspect much of what I wrote below is false._
Without taking proper care of a very deep neural network, gradients tend to suddenly become quite large or quite small. If the gradien... | https://www.lesswrong.com/posts/mdpyzdx9hMsjf9AKv/why-gradients-vanish-and-explode |
# What is the state of the ego depletion field?
It's been almost a half decade since the replication crisis in psychology broke, and one of the major casualties was the subfield of ego depletion. (I belive [this](https://journals.sagepub.com/doi/full/10.1177/1745691616652873) was the initial large scale meta-analysis ... | https://www.lesswrong.com/posts/xaMM9TsPRbhsZQj98/what-is-the-state-of-the-ego-depletion-field |
# Categorial preferences and utility functions
This post is motivated by [a recent post](https://www.lesswrong.com/posts/BydQtwfN97pFwEWtW/toy-model-piece-1-partial-preferences-revisited) of Stuart Armstrong on going from preferences to a utility function. It was originally planned as a comment, but seems to have deve... | https://www.lesswrong.com/posts/7ncFy84ReMFW7TDG6/categorial-preferences-and-utility-functions |
# Diana Fleischman and Geoffrey Miller - Audience Q&A
Cross-posted from Putanumonit.
* * *
This is the audience Q&A with [Diana Fleischman](https://www.dianafleischman.com/) and [Geoffrey Miller](https://www.primalpoly.com/) at the NYC Rationality meetup, following up on my own interview [which you can find here... | https://www.lesswrong.com/posts/LamLwo2XhdcRYTFSh/diana-fleischman-and-geoffrey-miller-audience-q-and-a |
# AI Safety Reading Group
If you are interested in AI Safety, come visit the AI Safety Reading Group.
The AI Safety reading group meets on Skype Wednesdays at 18:45 UTC, discussing new and old articles on different aspects of AI Safety. We start with a presentation round, then a summary of the article is presente... | https://www.lesswrong.com/posts/J55ak4zE2j2uDNevN/ai-safety-reading-group-1 |
# Could we solve this email mess if we all moved to paid emails?
Have you ever…
* Sent an email to someone in rationality and not heard back for many weeks (or more)?
* Avoided sending an email to someone because you wanted to spare their attention, despite thinking there was a fair chance they’d be genuinely in... | https://www.lesswrong.com/posts/tkuknrjYCbaDoZEh5/could-we-solve-this-email-mess-if-we-all-moved-to-paid |
# Adjectives from the Future: The Dangers of Result-based Descriptions
# Less Skeptical
Suppose your friend tells you he's on a weight-loss program. What do you think will happen in three months if he keeps on the weight-loss program? Will he lose weight?
If you're like me, you're thinking, "Of course. He is on a we... | https://www.lesswrong.com/posts/jnmkTMKaAExrPARmX/adjectives-from-the-future-the-dangers-of-result-based |
# Mesa-Optimizers and Over-optimization Failure (Optimizing and Goodhart Effects, Clarifying Thoughts - Part 4)
In the previous posts, I first outlined [Selection versus Control for Optimization](https://www.alignmentforum.org/posts/2neeoZ7idRbZf4eNC/re-introducing-selection-vs-control-for-optimization), then talked a... | https://www.lesswrong.com/posts/89YgRc3NevJEDEen5/mesa-optimizers-and-over-optimization-failure-optimizing-and |
# What explanatory power does Kahneman's System 2 possess?
In the 70s and 80s, Kahneman and Tversky did a bunch of pioneering research on heuristics and biases in human thought. Then, in _Thinking Fast and Slow,_ Kahneman divided human cognition into System 1 and System 2 - basically, System 1 applies quick heuristics... | https://www.lesswrong.com/posts/p5cMynf5de9FiyCxL/what-explanatory-power-does-kahneman-s-system-2-possess |
# A Primer on Matrix Calculus, Part 1: Basic review
Consider whether this story applies to you. You went through college and made it past linear algebra and multivariable calculus, and then began your training for deep learning. To your surprise, much of what they taught you in the previous courses is not very useful ... | https://www.lesswrong.com/posts/9L9XuXhLYBm47yYkf/a-primer-on-matrix-calculus-part-1-basic-review |
# Book Review: Secular Cycles
**I.**
There is a tide in the affairs of men. It cycles with a period of about three hundred years. During its flood, farms and businesses prosper, and great empires enjoy golden ages. During its ebb, war and famine stalk the land, and states collapse into barbarism.
![](http://slatesta... | https://www.lesswrong.com/posts/2weRdcvqANDq3zdPH/book-review-secular-cycles |
# Distance Functions are Hard
_\[Epistemic status: Describes a failed research approach I had a while ago, and my only purpose here is to warn people off from that way of thinking. Every now and then I see someone working on an AIS subproblem say "if only we had a distance function for things in domain X", and my intu... | https://www.lesswrong.com/posts/YuJNoCEgeWJfBtdtQ/distance-functions-are-hard-1 |
# Predicted AI alignment event/meeting calendar
Update 2020-06-21: Linda Linsefors made the [AI Safety Google Calendar](https://calendar.google.com/calendar?cid=Y2wxanRxMW80anNpcnJsMWdlaHE1a3BpanNAZ3JvdXAuY2FsZW5kYXIuZ29vZ2xlLmNvbQ). I published [How to make a predicted AI alignment event/meeting calendar](https://www... | https://www.lesswrong.com/posts/h8gypTEKcwqGsjjFT/predicted-ai-alignment-event-meeting-calendar |
# Subagents, trauma and rationality
**_\[Content note: discussion of trauma, child and sexual abuse, sexual violence, lack of self-worth, dissociation, PTSD, flashbacks, DID, personality disorders; some mildly graphic examples of abuse and trauma mentioned in text form\]_**
> I have spent over two years doing emotion... | https://www.lesswrong.com/posts/u5RLu5F3zKTB3Qjnu/subagents-trauma-and-rationality |
# Partial summary of debate with Benquo and Jessicata [pt 1]
_Note: I'll be trying not to engage too much with the object level discussion here – I think my marginal time on this topic is better spent thinking and writing longform thoughts. See [this comment](https://www.lesswrong.com/posts/FT9Lkoyd5DcCoPMYQ/partial-s... | https://www.lesswrong.com/posts/FT9Lkoyd5DcCoPMYQ/partial-summary-of-debate-with-benquo-and-jessicata-pt-1 |
# A Primer on Matrix Calculus, Part 2: Jacobians and other fun
I started this post thinking that I would write all the rules for evaluating Jacobians of neural network parameters in specific cases. But while this would certainly be useful for grokking deep learning papers, frankly it's difficult to write that in Latex... | https://www.lesswrong.com/posts/KKwv9kcQz29vqPLAD/a-primer-on-matrix-calculus-part-2-jacobians-and-other-fun |
# Clarifying some key hypotheses in AI alignment
We've created a diagram mapping out important **and** controversial hypotheses for AI alignment. We hope that this will help researchers identify and more productively discuss their disagreements.
## [Diagram](https://drive.google.com/open?id=1wI21XP-lRa6mi5h0dq_USooz... | https://www.lesswrong.com/posts/mJ5oNYnkYrd4sD5uE/clarifying-some-key-hypotheses-in-ai-alignment |
# What experiments would demonstrate "upper limits of augmented working memory?"
Wikipedia has this discussion of working-memory-as-ability-to-discern-relationships-simultaneously:
> Other have argued that working memory capacity is better characterized as "the ability to mentally form relations between elements, or ... | https://www.lesswrong.com/posts/8CggFo2xwTnZyFW9v/what-experiments-would-demonstrate-upper-limits-of-augmented |
# A Primer on Matrix Calculus, Part 3: The Chain Rule
This post concludes the subsequence on matrix calculus. Here, I will focus on an exploration of the chain rule as it's used for training neural networks. I initially planned to include Hessians, but perhaps for that we will have to wait.
* * *
Deep learning has t... | https://www.lesswrong.com/posts/u4DWp72GyLsQn24oR/a-primer-on-matrix-calculus-part-3-the-chain-rule |
# How can you use music to boost learning?
I often find that I am able to appreciate the beauty of a subject more while listening to music (especially instrumental music). Hearing the notes while I think about the topic helps creates a lot of subconscious connections with the material, solidifying what I am learning a... | https://www.lesswrong.com/posts/YpzNrhAAfnsX24YtZ/how-can-you-use-music-to-boost-learning |
# Problems in AI Alignment that philosophers could potentially contribute to
(This was originally a comment that I wrote as a follow up to [my question](https://ea.greaterwrong.com/posts/oPGJrqohDqT8GZieA/ask-me-anything/comment/cL3KFfJwtHZKeBPGH) for William MacAskill's AMA. I'm moving it since it's perhaps more on-t... | https://www.lesswrong.com/posts/rASeoR7iZ9Fokzh7L/problems-in-ai-alignment-that-philosophers-could-potentially |
# Inspection Paradox as a Driver of Group Separation
> _Have you ever noticed that anybody driving slower than you is an idiot, and anyone going faster than you is a maniac?_ \-\- George Carlin
The [Inspection Paradox](http://allendowney.blogspot.com/2015/08/the-inspection-paradox-is-everywhere.html), where the repor... | https://www.lesswrong.com/posts/M52wqeyCeLC766Pn9/inspection-paradox-as-a-driver-of-group-separation |
# Neural Nets in Python 1
Introduction
============
This post is an attempt to explain how to write a neural network in Python using numpy. I am obviously not the first person to do this. Almost all of the code is here adapted from Michael Nielsen's fantastic online book [Neural Networks and Deep Learning](http:/... | https://www.lesswrong.com/posts/HHWubYS7Mf3nu7XT6/neural-nets-in-python-1 |
# "Can We Survive Technology" by von Neumann
> "The great globe itself" is in a rapidly maturing crisis —a crisis attributable to the fact that the environment in which technological progress must occur has become both undersized and underorganized. To define the crisis with any accuracy, and to explore possibilities ... | https://www.lesswrong.com/posts/J6i2FWXQSK4ybGHWT/can-we-survive-technology-by-von-neumann |
# Raph Koster on Virtual Worlds vs Games (notes)
Raph Koster is a game designer who's worked on old-school MUDs, Ultima Online, and Star Wars Galaxies among others. [His blog](https://www.raphkoster.com/) is a treasure trove of information on game design, and online community building.
The vibe I get is very sequence... | https://www.lesswrong.com/posts/uicb62iovv6HC5g7E/raph-koster-on-virtual-worlds-vs-games-notes |
# Goodhart's Curse and Limitations on AI Alignment
I believe that most existing proposals for aligning AI with human values are unlikely to succeed in the limit of optimization pressure due to Goodhart's curse. I believe this strongly enough that it continues to surprise me a bit that people keep working on things tha... | https://www.lesswrong.com/posts/NqQxTn5MKEYhSnbuB/goodhart-s-curse-and-limitations-on-ai-alignment |
# Unstriving
[Cross-posted from Putanumonit](https://putanumonit.com/2019/08/15/unstriving/)
* * *
Successometer
-------------
Since I was a kid, I have built my self-esteem on a feeling of “forthcoming greatness”. Whatever I actually accomplished I never paused to be proud of, and I never sweated the failures eith... | https://www.lesswrong.com/posts/o2cwMtZpwo3zhbFEd/unstriving |
# Classifying specification problems as variants of Goodhart's Law
There are a few different classifications of safety problems, including the [Specification, Robustness and Assurance (SRA) taxonomy](https://medium.com/@deepmindsafetyresearch/building-safe-artificial-intelligence-52f5f75058f1) and the [Goodhart's Law ... | https://www.lesswrong.com/posts/yXPT4nr4as7JvxLQa/classifying-specification-problems-as-variants-of-goodhart-s |
# Do We Change Our Minds Less Often Than We Think?
In "[We Change Our Minds Less Often Than We Think](https://www.lesswrong.com/posts/buixYfcXBah9hbSNZ/we-change-our-minds-less-often-than-we-think)", Eliezer quotes a study:
> Over the past few years, we have discreetly approached colleagues faced with a choice betwee... | https://www.lesswrong.com/posts/FSvRW8wxmriqcYD4t/do-we-change-our-minds-less-often-than-we-think |
# Negative "eeny meeny miny moe"
As a kid, I learned the rhyme as:
> Eeny, meeny, miny, moe,
> Catch a tiger by the toe.
> If he hollers, let him go,
> Out goes Y, O, U!
Since kids can't predict where it will end, and adults are not supposed to try, it's a reasonably fair way of drawing lots.
At times I've heard... | https://www.lesswrong.com/posts/frngWKRpRdHPXxnCC/negative-eeny-meeny-miny-moe |
# Self-supervised learning & manipulative predictions
**Abstract:** I wrote recently about [Self-Supervised Learning and AGI Safety](https://www.lesswrong.com/posts/EMZeJ7vpfeF4GrWwm/self-supervised-learning-and-agi-safety) in general. This post discusses one potential failure mode in more detail. Take a self-supervis... | https://www.lesswrong.com/posts/L3Ryxszc3X2J7WRwt/self-supervised-learning-and-manipulative-predictions |
# Has Moore's Law actually slowed down?
Moore's Law has been notorious for spurring a bunch of separate observations that are all covered under the umbrella of "Moore's law." But as far as I can tell, the real Moore's law is a fairly narrow prediction, which is that the transistor count on CPU chips will double approx... | https://www.lesswrong.com/posts/motErF8bKhbtDcLut/has-moore-s-law-actually-slowed-down |
# What are the reasons to *not* consider reducing AI-Xrisk the highest priority cause?
ETA: I'll be adding things to the list that I think belong there.
I'm assuming a high level of credence in classic utilitarianism, and that AI-Xrisk is significant (e.g. roughly >10%), and timelines are not long (e.g. >50% ASI in <... | https://www.lesswrong.com/posts/KEjJRYSPMESX48mDd/what-are-the-reasons-to-not-consider-reducing-ai-xrisk-the |
# What authors consistently give accurate pictures of complex topics they discuss?
When I read nonfiction, I often feel like it's a waste of time because I'm not very confident the author is portraying things accurately/non-misleadingly, and don't often have time or expertise to check (though I'd potentially be in doi... | https://www.lesswrong.com/posts/q4D248oZFwBJTaZLs/what-authors-consistently-give-accurate-pictures-of-complex |
# GPT-2: 6-Month Follow-Up
Linkpost for [GPT-2 6 Month Follow-Up](https://openai.com/blog/gpt-2-6-month-follow-up/).
Some highlights:
- 700+ M parameter model is being released
- Several other groups have reproduced similar models
- In detecting synthesized text, "current ML-based methods only achieve low to mid–90s ... | https://www.lesswrong.com/posts/t9MP6ZPqhTTvjssuB/gpt-2-6-month-follow-up |
# Three Levels of Motivation
Disclaimer: The information herein is not at all new. I wrote this a year ago, but when I wanted to link to it recently I found out it wasn't in existence on LW (except as a link post which isn't very satisfactory). I decided to polish it up and post it here.
Epistemic Status
------... | https://www.lesswrong.com/posts/Pq2XhSGbQWF8Z7h8y/three-levels-of-motivation-1 |
# Paradoxical Advice Thread
_(Edited to be more clear)_
One can find plenty of witty advice catch phrases that appear to be in direct conflict with each other: "Early bird gets the work, but the second mouse gets the cheese" (from a comment below)
You can also find this with things that are more detailed frames than... | https://www.lesswrong.com/posts/nRZyYZmYdTDfCThok/paradoxical-advice-thread |
# Two senses of “optimizer”
The word “optimizer” can be used in at least two different ways.
First, a system can be an “optimizer” in the sense that it is solving a computational optimization problem. A computer running a linear program solver, a SAT-solver, or gradient descent, would be an example of a system that i... | https://www.lesswrong.com/posts/rvxcSc6wdcCfaX6GZ/two-senses-of-optimizer |
# Call for contributors to the Alignment Newsletter
**TL;DR:** I am looking for (possibly paid) contributors to write summaries and opinions for the Alignment Newsletter. This is currently experimental, but I estimate ~80% chance that it will become long-term, and so I’m looking for people who are likely to contribute... | https://www.lesswrong.com/posts/cejX4S3Dex3C2dt79/call-for-contributors-to-the-alignment-newsletter |
# Announcement: Writing Day Today (Thursday)
The MIRI Summer Fellows Program is having a writing day, where the participants are given a whole day to write whatever LessWrong / AI Alignment Forum posts that they like. This is to practice the skills needed for [forum participation as a research strategy](https://www.le... | https://www.lesswrong.com/posts/aJGXWHnYTsWwAiwf6/announcement-writing-day-today-thursday |
# Markets are Universal for Logical Induction
Background
----------
[Logical Induction](https://arxiv.org/pdf/1609.03543.pdf) is the best framework currently available for thinking about logical uncertainty - i.e. the “probability” that the [twin primes conjecture](https://en.wikipedia.org/wiki/Twin_prime#History) is... | https://www.lesswrong.com/posts/WmNeCipNwg9CmGy3T/markets-are-universal-for-logical-induction |
# Body Alignment & Balance. Our Midline Anatomy & the Median Plane.
Alignment & Balance of the Human Body.
======================================
*Epistemic status: An introduction post for my Base-Line Theory of Human Health and Movement. Facts and definitions first ...*
Introduction.
-------------
When referrin... | https://www.lesswrong.com/posts/AZC55cWzX6aFngnRN/body-alignment-and-balance-our-midline-anatomy-and-the |
# Implications of Quantum Computing for Artificial Intelligence Alignment Research
_\[Crossposted from [arXiv]( https://arxiv.org/abs/1908.07613 )\]_
**ABSTRACT:** We introduce a heuristic model of Quantum Computing and apply it to argue that a deep understanding of quantum computing is unlikely to be helpful to addr... | https://www.lesswrong.com/posts/ZkgqsyWgyDx4ZssqJ/implications-of-quantum-computing-for-artificial |
# [AN #62] Are adversarial examples caused by real but imperceptible features?
Find all Alignment Newsletter resources [here](http://rohinshah.com/alignment-newsletter/). In particular, you can [sign up](http://eepurl.com/dqMSZj), or look through this [spreadsheet](https://docs.google.com/spreadsheets/d/1PwWbWZ6FPqAgZ... | https://www.lesswrong.com/posts/NTwA3J99RPkgmp6jh/an-62-are-adversarial-examples-caused-by-real-but |
# Computational Model: Causal Diagrams with Symmetry
Consider the following program:
```
f(n):
if n == 0:
return 1
return n * f(n-1)
```
Let’s think about the process by which this function is evaluated. We want to sketch out a [causal DAG](https://www.lesswrong.com/posts/hzuSDMx7pd2uxFc5w/causal-diagr... | https://www.lesswrong.com/posts/mZy6AMgCw9CPjNCoK/computational-model-causal-diagrams-with-symmetry |
# Intentional Bucket Errors
I want to illustrate a research technique that I use sometimes. (My actual motivation for writing this is to make it so that I don't feel as much like I need to defend myself when I use this technique.) I am calling it intentional bucket errors after a CFAR concept called [bucket errors](ht... | https://www.lesswrong.com/posts/ZCfwKjQhAiDogorj8/intentional-bucket-errors |
# Embedded Naive Bayes
Suppose we have a bunch of earthquake sensors spread over an area. They are not perfectly reliable (in terms of either false positives or false negatives), but some are more reliable than others. How can we aggregate the sensor data to detect earthquakes?
A “naive” seismologist without any stat... | https://www.lesswrong.com/posts/bKa3JEQYv9sSJ3d2h/embedded-naive-bayes |
# Time Travel, AI and Transparent Newcomb
_Epistemic status: has "time travel" in the title._
Let's suppose, for the duration of this post, that the local physics of our universe allows for time travel. The obvious question is: how are paradoxes prevented?
We may not have any idea how paradoxes are prevented, but pr... | https://www.lesswrong.com/posts/o5wp9x55tpPizkdGk/time-travel-ai-and-transparent-newcomb |
# Logical Counterfactuals and Proposition graphs, Part 1
Within this sequence of posts I will outline a procedure for logical counterfactuals based on something similar to proof length. In this post I present a reimagining of propositional logic in which proving a theorem is taking a walk around a graph of equivalent ... | https://www.lesswrong.com/posts/Pxvq2RMAKCuY6SHm9/logical-counterfactuals-and-proposition-graphs-part-1 |
# Why so much variance in human intelligence?
_Epistemic status: Practising thinking aloud. There might be an important question here, but I might be making a simple error._
There is a lot of variance in general competence between species. Here is the standard Bostrom/Yudkowsky graph to display this notion.
, "Why I Am Not A Technocrat", a major focus of which is his differences with the rationalist community.
I feel like I've read a decent number of... | https://www.lesswrong.com/posts/6pZzsSFkwkz74CBrr/response-to-glen-weyl-on-technocracy-and-the-rationalist |
# Towards a mechanistic understanding of corrigibility
## Acceptability
To be able to use something like [relaxed adversarial training](https://ai-alignment.com/training-robust-corrigibility-ce0e0a3b9b4d) to verify a model, a necessary condition is having a good notion of _acceptability._ [Paul Christiano describes t... | https://www.lesswrong.com/posts/BKM8uQS6QdJPZLqCr/towards-a-mechanistic-understanding-of-corrigibility |
# Logical Optimizers
EDIT: This idea is unsafe unless you have X and only X optimization. If an early logical optimiser produces a paperclip maximiser, then the paperclip maximiser will play along giving good solutions to most logical optimisation problems, but hiding nasty surprises where we are likely to find them... | https://www.lesswrong.com/posts/ENEYeeyPXyuEQdtZy/logical-optimizers |
# Deconfuse Yourself about Agency
_This post is a result of numerous discussions with other participants and organizers of the MIRI Summer Fellows Program 2019._
-----------------------
I recently (hopefully) dissolved some of my confusion about agency. In the first part of the post, I describe a concept that I bel... | https://www.lesswrong.com/posts/ZigRhB4pAGdr6beQh/deconfuse-yourself-about-agency |
# Thoughts from a Two Boxer
I'm writing this for blog day at MSFP. I thought about a lot of things here like category theory, the [1-2-3 conjecture](https://arxiv.org/abs/1211.5122) and Paul Christiano's agenda. I want to start by thanking everyone for having me and saying I had a really good time. At this point I int... | https://www.lesswrong.com/posts/9QztnkMiKJ7jYZhL8/thoughts-from-a-two-boxer |
# Analysis of a Secret Hitler Scenario
Secret Hitler is a social deception game in the tradition of mafia, the resistance and Avalon [^1]. You can read the rules [here](https://secrethitler.com/assets/Secret_Hitler_Rules.pdf) if you aren't familiar. I haven't played social deception games regularly since 2016 but in m... | https://www.lesswrong.com/posts/C8geRCXF54FsyKydL/analysis-of-a-secret-hitler-scenario-1 |
# The Commitment Races problem
_\[Epistemic status: Strong claims vaguely stated and weakly held. I expect that writing this and digesting feedback on it will lead to a much better version in the future. EDIT: So far this has stood the test of time. EDIT: As of September 2020 I think this is one of the most import... | https://www.lesswrong.com/posts/brXr7PJ2W4Na2EW2q/the-commitment-races-problem |
# Does Agent-like Behavior Imply Agent-like Architecture?
This is not a well-specified question. I don't know what "agent-like behavior" or "agent-like architecture" should mean. Perhaps the question should be "Can you define the fuzzy terms such that 'Agent-like behavior implies agent-like architecture' is true, usef... | https://www.lesswrong.com/posts/osxNg6yBCJ4ur9hpi/does-agent-like-behavior-imply-agent-like-architecture |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.