text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Discriminating Behaviorally Identical Classifiers: a model problem for applying interpretability to scalable oversight
In a new preprint, [Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models](https://arxiv.org/abs/2403.19647), my coauthors and I introduce a technique, Spar... | https://www.lesswrong.com/posts/s7uD3tzHMvD868ehr/discriminating-behaviorally-identical-classifiers-a-model |
# [Fiction] A Confession
This morning while taking the LIRR to the city I performed first aid on a man who had been shot through the window of my carriage.
“Is he going to die?” his girlfriend asked me.
“We’re all going to die.”
A long pause. “I mean—is he going to die right now?”
“Probably not.” Probably he didn’... | https://www.lesswrong.com/posts/tu4qwFYJeDDER8bch/fiction-a-confession |
# Blessed information, garbage information, cursed information
*This post is also available on* [*my substack*](https://tailcalled.substack.com/p/blessed-information-garbage-information)*. Thanks to Justis Mills for editing and feedback.*
Imagine that you're a devops engineer who has been tasked with solving an incid... | https://www.lesswrong.com/posts/gjWdzvtFtFZ2TEcjs/blessed-information-garbage-information-cursed-information |
# I'm open for projects (sort of)
I left Google a month ago, and right now don't work. Writing this post in case anyone has interesting ideas what I could do. This isn't an "urgently need help" kind of thing - I have a little bit of savings, right now planning to relax some more weeks and then go into some solo softwa... | https://www.lesswrong.com/posts/DfqxDezqCeFJPjYsL/i-m-open-for-projects-sort-of |
# A Review of In-Context Learning Hypotheses for Automated AI Alignment Research
*This project has been completed as part of the Mentorship in Alignment Research Students (MARS London) programme under the supervision of Bogdan-Ionut Cirstea, on investigating the promise of automated AI alignment research. I would like... | https://www.lesswrong.com/posts/GPcwP8pgyPFPwvi2h/a-review-of-in-context-learning-hypotheses-for-automated-ai |
# LessOnline Festival Updates Thread
This is a thread for updates about the upcoming [**LessOnline**](https://less.online/) festival. I (Ben) will be posting bits of news and thoughts, and you're also welcome to make suggestions or ask questions.
If you'd like to hear about new updates, you can use LessWrong's "Subsc... | https://www.lesswrong.com/posts/MmWziepD8DDauSide/lessonline-festival-updates-thread |
# hydrogen tube transport
Elon Musk's [Hyperloop proposal](https://en.wikipedia.org/wiki/Hyperloop) had substantial public interest. With various initial Hyperloop projects now having failed, I thought some people might be interested in a high-speed transportation system that's...perhaps not "practical" per se, but at... | https://www.lesswrong.com/posts/foseau7Eqg22mCkpY/hydrogen-tube-transport |
# Backyard Office
In 2020 I renovated the small building in our backyard which had [fallen into disrepair](https://www.jefftk.com/p/outbuilding-thoughts). It was zoned for use as a home office, and had electric but not plumbing. I wrote about how I was [thinking about insulating it](https://www.jefftk.com/p/estimating... | https://www.lesswrong.com/posts/wonC6LQArRcFXgebZ/backyard-office |
# Cohesion and business problems
In software development there is a concept called [cohesion](https://en.wikipedia.org/wiki/Cohesion_(computer_science)).
It works like this. Suppose you have the following functions:[^dplw8r0y7m]
function getArea(radius) { ... }
function getCircumference(radius) { ... }
... | https://www.lesswrong.com/posts/YCFsfQzAnXW4GEKS5/cohesion-and-business-problems |
# Experiment on repeating choices
People behave differently from one another on all manner of axes, and each person is usually pretty consistent about it. For instance:
* how much to spend money
* how much to worry
* how much to listen vs. speak
* how much to jump to conclusions
* how much to work
* how p... | https://www.lesswrong.com/posts/RPdG2fSiSPLkCSPjg/experiment-on-repeating-choices |
# If digital goods in virtual worlds increase GDP, do we actually become richer?
Noah Smith, in [this article](https://www.noahpinion.blog/p/the-metaverse-and-near-infinite-economic), argues that the Metaverse could enable economic growth to increase a lot and sharply decouple itself from real-world resource usage. By... | https://www.lesswrong.com/posts/beFF5GyBy4FntYFRQ/if-digital-goods-in-virtual-worlds-increase-gdp-do-we |
# What's up with all the non-Mormons? Weirdly specific universalities across LLMs
**tl;dr:** Recently reported GPT-J experiments \[[1](https://www.lesswrong.com/posts/c6uTNm5erRrmyJvvD/mapping-the-semantic-void-strange-goings-on-in-gpt-embedding) [2](https://www.lesswrong.com/posts/hincdPwgBTfdnBzFf/mapping-the-semant... | https://www.lesswrong.com/posts/CNPvESPru3XNqsw7A/what-s-up-with-all-the-non-mormons-weirdly-specific |
# How to Model the Future of Open-Source LLMs?
I previously expected open-source LLMs to lag far behind the frontier because they're very expensive to train and naively it doesn't make business sense to spend on the order of $10M to (soon?) $1B to train a model only to give it away for free.
But this has been repeate... | https://www.lesswrong.com/posts/vsaMKdJNQHgSxWy2a/how-to-model-the-future-of-open-source-llms |
# CTMU insight: maybe consciousness *can* affect quantum outcomes?
From one of [justinpombrio’s comments](https://www.lesswrong.com/posts/pzmRDnoi4mNtqu6Ji/the-cognitive-theoretic-model-of-the-universe-a-partial?commentId=NZrciirsgs3vFyFBr) on [Jessica Taylor’s review of the CTMU](https://www.lesswrong.com/posts/pzmRD... | https://www.lesswrong.com/posts/F2iwdBudEpERC8yNc/ctmu-insight-maybe-consciousness-can-affect-quantum-outcomes |
# What is the best way to talk about probabilities you expect to change with evidence/experiments?
I was thinking about my p(doom) in the next 10 years and came up with something around 6%[^wdzr3vimkyj]. However that involves lots of current unknowns to me, like the nature of current human knowledge production (and th... | https://www.lesswrong.com/posts/v2FReduuvLWZA3jyr/what-is-the-best-way-to-talk-about-probabilities-you-expect |
# Daniel Dennett has died (1942-2024)
> Daniel Dennett, professor emeritus of philosophy at Tufts University, well-known for his work in philosophy of mind and a wide range of other philosophical areas, has died.
>
> [Professor Dennett](https://as.tufts.edu/philosophy/people/faculty/daniel-dennett) wrote extensively ... | https://www.lesswrong.com/posts/X5bXnA7WHopGoMH4X/daniel-dennett-has-died-1942-2024 |
# [Summary] Progress Update #1 from the GDM Mech Interp Team
Introduction
============
This is a progress update from the Google DeepMind mechanistic interpretability team, inspired by the Anthropic team’s [excellent monthly updates](https://transformer-circuits.pub/)! Our goal was to write-up a series of snippets, c... | https://www.lesswrong.com/posts/HpAr8k74mW4ivCvCu/summary-progress-update-1-from-the-gdm-mech-interp-team |
# [Full Post] Progress Update #1 from the GDM Mech Interp Team
*This is a series of snippets about the Google DeepMind mechanistic interpretability team's research into Sparse Autoencoders, that didn't meet our bar for a full paper. Please start at* [*the summary post*](https://www.alignmentforum.org/posts/HpAr8k74mW4... | https://www.lesswrong.com/posts/C5KAZQib3bzzpeyrg/full-post-progress-update-1-from-the-gdm-mech-interp-team |
# Inducing Unprompted Misalignment in LLMs
Emergent Instrumental Reasoning Without Explicit Goals
======================================================
**TL;DR**: LLMs can act and scheme without being told to do so. This is bad.
*Produced as part of Astra Fellowship - Winter 2024 program, mentored by Evan Hubing... | https://www.lesswrong.com/posts/ukTLGe5CQq9w8FMne/inducing-unprompted-misalignment-in-llms |
# Thoughts on seed oil
A friend has spent the last three years hounding me about seed oils. Every time I thought I was safe, he’d wait a couple months and renew his attack:
> *“When are you going to write about seed oils?”*
> *“Did you know that seed oils are why there’s so much {obesity, heart disease, diabetes, in... | https://www.lesswrong.com/posts/DHkkL2GxhxoceLzua/thoughts-on-seed-oil |
# Past Tense Features
TLDR
----
I find past tense features in pythia-70m using a templated dataset. My high-level steps are:
1. Creating a templated dataset that indicates past tense through a past progressive clause
2. Finding subsets of features that recover the original model performance with attribution patchi... | https://www.lesswrong.com/posts/KE6X8peAKutYNXnx5/past-tense-features |
# A couple productivity tips for overthinkers
1\. If you find that you’re reluctant to permanently give up on to-do list items, “deprioritize” them instead
-------------------------------------------------------------------------------------------------------------
I *hate* the idea of deciding that something on my t... | https://www.lesswrong.com/posts/ZN6L5ysKd35FEyGr6/a-couple-productivity-tips-for-overthinkers |
# Good Bings copy, great Bings steal
Stop me if you've heard this one before:
> LLMs may produce a lot of seemingly original text, but really, they can never equate a human's output, because they can only *remix* material from their own training set, never *create* new knowledge.
Ok, no one stopped me, but I imagine... | https://www.lesswrong.com/posts/r587cE9sW6fBhtXSv/good-bings-copy-great-bings-steal |
# The losing identity of Twitter
What’s Twitter for you?
That's a long-lasting trend I often see on my feed when people praise the blue bird for getting them a job, introducing them to new people, investors, and all this and that.
What about me? I just wanted to get into [dribbble](https://dribbble.com/) — the then ... | https://www.lesswrong.com/posts/Eu6Rsc6bThkAosFCY/the-losing-identity-of-twitter |
# Transfer Learning in Humans
__I examine the literature on transfer learning in humans. [Far
transfer](#Far_Transfer) is difficult to achieve, best
candidate interventions are to [practice at the edge of one's
ability and make many mistakes](#Error_Management_Training),
[evaluate mistakes](#Inducing_Transfer) after o... | https://www.lesswrong.com/posts/QTTCRytvyFteJgPwg/transfer-learning-in-humans |
# Time complexity for deterministic string machines
This was a project conducted during [MATS 5.0](https://www.matsprogram.org/) under the mentorship of [Vanessa Kosoy](https://www.lesswrong.com/users/vanessa-kosoy) and supported by a grant from [BERI](https://existence.org/). It builds off the [String Machines](https... | https://www.lesswrong.com/posts/zy2AECRAi8Nuu5XMk/time-complexity-for-deterministic-string-machines |
# Goal oriented cognition in "a single forward pass"
The below is me (habryka) and dxu talking about a shortform that dxu had published a few months ago, going into the relationship between goal-oriented cognition and the myopic nature of current large language model training setups. Some key quotes if you don't want ... | https://www.lesswrong.com/posts/H6rc8xFbdKYw39ihu/goal-oriented-cognition-in-a-single-forward-pass |
# Motivation gaps: Why so much EA criticism is hostile and lazy
*Disclaimer: While I criticize several EA critics in this article, I am myself on the EA-skeptical side of things (especially on AI risk).*
### **Introduction**
I am a proud critic of effective altruism, and in particular a critic of AI existential ris... | https://www.lesswrong.com/posts/ojPJYci3AdgdHsKY8/motivation-gaps-why-so-much-ea-criticism-is-hostile-and-lazy |
# On Llama-3 and Dwarkesh Patel’s Podcast with Zuckerberg
It was all quiet. Then it wasn’t.
[Note the timestamps on both of these](https://twitter.com/tszzl/status/1781043498801893827).
[
**EDIT: I somehow missed that John Wentworth and David Lorell** ~~**are also in the middle of a sequence**~~ **have written one post on this same topic** [**here**](https://www.lesswrong.com/posts/keGhMbnLNdTAHa4bv/generalized-stat-mech-the-boltzmann-approach)**.** ~~... | https://www.lesswrong.com/posts/wbnWNSCyjFKyknvBy/forget-everything-statistical-mechanics-part-1 |
# Priors and Prejudice
I
-
Imagine an alternate version of the Effective Altruism movement, whose early influences came from socialist intellectual communities such as the[ Fabian Society](https://forum.effectivealtruism.org/posts/oDkfxGrT9RfadCNFc/the-fabian-society-was-weirdly-similar-to-the-ea-movement), as oppose... | https://www.lesswrong.com/posts/sKKxuqca9uhpFSvgq/priors-and-prejudice |
# AI Regulation is Unsafe
Concerns over AI safety and calls for government control over the technology are highly correlated but they should not be.
There are two major forms of AI risk: misuse and misalignment. Misuse risks come from humans using AIs as tools in dangerous ways. Misalignment risks arise if AIs take t... | https://www.lesswrong.com/posts/3LuZm3Lhxt6aSpMjF/ai-regulation-is-unsafe |
# Measuring Coherence and Goal-Directedness in RL Policies
*This post was produced as part of the* [*Astra Fellowship*](https://www.constellation.org/programs/astra-fellowship) *under the Winter 2024 Cohort, mentored by Richard Ngo. *
*Epistemic status: relatively confident in the overall direction of this post, but ... | https://www.lesswrong.com/posts/g5XgjY6x9dz47LgAz/measuring-coherence-and-goal-directedness-in-rl-policies |
# Funny Anecdote of Eliezer From His Sister
This comes from a podcast called 18Forty, of which the main demographic of Orthodox Jews. Eliezer's sister (Hannah) came on and talked about her Sheva Brachos, which is essentially the marriage ceremony in Orthodox Judaism. People here have likely not seen it, and I thought ... | https://www.lesswrong.com/posts/C7deNdJkdtbzPtsQe/funny-anecdote-of-eliezer-from-his-sister |
# Thoughts on Zero Points
There’s a cool concept I’ve been thinking about. I first heard of it when reading Jesse Schell’s book “The Art of Game Design”. (Fun fact: Jesse Schell was my professor’s professor, aka my grand-professor.)
Then I heard of it again in the LessWrong post “[Choosing the Zero Point](https://www... | https://www.lesswrong.com/posts/RYCvRwwKumBTPrhax/thoughts-on-zero-points |
# LW Frontpage Experiments! (aka "Take the wheel, Shoggoth!")
Update: June 20th
-----------------
After a few rounds of adjustments and careful examination of the data, we've decided to make the [Enriched tab be the default for all logged-in users](https://www.lesswrong.com/posts/TDMKch5qzuaac5LFF/enriched-tab-is-now... | https://www.lesswrong.com/posts/F6pf38EQMxtMA5J36/lw-frontpage-experiments-aka-take-the-wheel-shoggoth |
# Rejecting Television
I didn’t use to be, but now I’m part of the [2%](https://www.statista.com/statistics/243789/number-of-tv-households-in-the-us/) of [U.S. households](https://www.google.com/search?q=how+many+households+in+the+us&rlz=1C1CHBF_enUS924US924&oq=how+many+households+&gs_lcrp=EgZjaHJvbWUqBwgAEAAYgAQyBwgA... | https://www.lesswrong.com/posts/JE37rMdXqhxSaoDDC/rejecting-television |
# ProLU: A Nonlinearity for Sparse Autoencoders
Abstract
========
This paper presents $\text{ProLU}$, an alternative to $\text{ReLU}$ for the activation function in sparse autoencoders that produces a pareto improvement over both standard sparse autoencoders trained with an L1 penalty and sparse autoencoders trained ... | https://www.lesswrong.com/posts/HEpufTdakGTTKgoYF/prolu-a-nonlinearity-for-sparse-autoencoders |
# Vector Planning in a Lattice Graph
You want to get to your sandwich:
Well, that’s easy. Apparently we are in some kind of grid world, which is presented to us in the form of a lattice graph, where eac... | https://www.lesswrong.com/posts/w5DPADKQw2KdqujgE/vector-planning-in-a-lattice-graph |
# Dequantifying first-order theories
The [Löwenheim–Skolem theorem](https://en.wikipedia.org/wiki/L%C3%B6wenheim%E2%80%93Skolem_theorem) implies, among other things, that any first-order theory whose symbols are countable, and which has an infinite model, has a countably infinite model. This means that, in attempting ... | https://www.lesswrong.com/posts/c6oLNbKNDtGE5xpNL/dequantifying-first-order-theories-1 |
# On what research policymakers actually need
I saw this guest post on the Slow Boring substack, by a former senior US government official, and figured it might be of interest here. The post's original title is "The economic research policymakers actually need", but it seemed to me like the post could be applied just ... | https://www.lesswrong.com/posts/y4rz6BFENLBThxrHm/on-what-research-policymakers-actually-need |
# Book review: Deep Utopia
Book review: Deep Utopia: Life and Meaning in a Solved World, by Nick
Bostrom.
Bostrom's previous book,
[Superintelligence](https://bayesianinvestor.com/blog/index.php/2014/07/28/superintelligence/),
triggered expressions of concern. In his latest work, he describes his
hopes for the ... | https://www.lesswrong.com/posts/AfABcGZshpyx2nxiZ/book-review-deep-utopia |
# (When) Should you work through the night when inspiration strikes you?
A lot of the time, I'm not very motivated to work, at least on particular projects. Sometimes, I feel very inspired and motivated to work on a particular project that I usually don't feel (as) motivated to work on. Sometimes, this happens in the ... | https://www.lesswrong.com/posts/99P2GwacTwp2mjhxq/when-should-you-work-through-the-night-when-inspiration |
# Simple probes can catch sleeper agents
*This is a link post for the Anthropic Alignment Science team's first "Alignment Note" blog post. We expect to use this format to showcase early-stage research and work-in-progress updates more in the future.*
[Twitter thread here.](https://twitter.com/AnthropicAI/status/17829... | https://www.lesswrong.com/posts/gknc6NWCNuTCe8ekp/simple-probes-can-catch-sleeper-agents-1 |
# On Minicircle
[Max Berry has analyzed](https://twitter.com/yashgaroth/status/1782497300876591431) Minicircle's follistatin gene therapy, and I agree with his conclusion that it is unlikely to be effective. There are several aspects of their design that could be improved (in particular, using a more efficient lipid n... | https://www.lesswrong.com/posts/R5EJn3TFpHDQBW3Af/on-minicircle |
# Let's Design A School, Part 1
The American school system, grades K-12, leaves much to be desired.
While its [flaws](https://www.wesjones.com/gatto1.htm) [are](https://en.wikipedia.org/wiki/The_Case_Against_Education) [legion](https://www.lesswrong.com/posts/fLPcQ5oqJ8oFxRwwL/the-case-against-education), this post i... | https://www.lesswrong.com/posts/uAQxLJtGrRQg3LgqA/let-s-design-a-school-part-1 |
# Is there software to practice reading expressions?
I took the [Reading the Mind in the Eyes Test](https://s3.amazonaws.com/he-assets-prod/interactives/233_reading_the_mind_through_eyes/Launch.html) test today. I got 27/36. [Jessica Livingston](https://www.paulgraham.com/jessica.html) got 36/36.
Reading expressions ... | https://www.lesswrong.com/posts/hTgWygNXbEo4mAgXM/is-there-software-to-practice-reading-expressions |
# Examples of Highly Counterfactual Discoveries?
The history of science has tons of examples of the same thing being discovered multiple time independently; wikipedia has a whole list of examples [here](https://en.wikipedia.org/wiki/List_of_multiple_discoveries). If your goal in studying the history of science is to e... | https://www.lesswrong.com/posts/csHstEPagqs8wChhh/examples-of-highly-counterfactual-discoveries |
# AI Generated Music as a Method of Installing Essential Rationalist Skills
On April Fools the LW team released an [album](https://music.youtube.com/playlist?list=OLAK5uy_mJVqWjzbOu3O5yegpCHim7d1UwU5waLnU) under the name of the Fooming Shoggoths. Ever since the amount that I think about rationality has skyrocketed.
T... | https://www.lesswrong.com/posts/LSDtWbE6acmff6Tig/ai-generated-music-as-a-method-of-installing-essential |
# 1-page outline of Carlsmith's otherness and control series
*Joe’s summary is *[*here*](https://joecarlsmith.com/2024/01/02/otherness-and-control-in-the-age-of-agi)*, these are my condensed takeaways in my own words. All links in this section are to the* [*essays*](https://www.lesswrong.com/s/BbAvHtorCZqp97X9W)*. *
... | https://www.lesswrong.com/posts/qxakrNr3JEoRSZ8LE/1-page-outline-of-carlsmith-s-otherness-and-control-series |
# Changes in College Admissions
This post brings together various questions about the college application process, as well as practical considerations of where to apply and go. We are seeing some encouraging developments, but mostly the situation remains rather terrible for all concerned.
#### Application Strategy an... | https://www.lesswrong.com/posts/PTC7bZdZoqbCcAshW/changes-in-college-admissions |
# Magic by forgetting
*Epistemic – this post is more suitable for LW as it was 10 years ago*
***Thought experiment with curing a disease by forgetting***
Imagine I have a bad but rare disease X. I may try to escape it in the following way:
1\. I enter the blank state of mind and forget that I had X.
2\. Now I in s... | https://www.lesswrong.com/posts/DNzzfBAZx8Laua4MB/magic-by-forgetting |
# Betadine oral rinses for covid and other viral infections
Before we get started, this is your quarterly reminder that I have no medical credentials and my highest academic credential is a BA in a different part of biology (with a double major in computer science). In a world with a functional medical system no one w... | https://www.lesswrong.com/posts/ZNu7rQLnggos5wZt3/betadine-oral-rinses-for-covid-and-other-viral-infections |
# This is Water by David Foster Wallace
*Note: It seems like great essays should go here and be fed through the standard LessWrong algorithm. There is possibly a copyright issue here, but we aren't making any money off it either. What follows is a full copy of "This is Water" by David Foster Wallace his 2005 commencem... | https://www.lesswrong.com/posts/2x8EJnEu4JZMiNbxd/this-is-water-by-david-foster-wallace |
# The Inner Ring by C. S. Lewis
*Note: In* [*@Nathan Young*](https://www.lesswrong.com/users/nathan-young?mention=user)*'s* [*words*](https://www.lesswrong.com/posts/2x8EJnEu4JZMiNbxd/this-is-water-by-david-foster-wallace) *"It seems like great essays should go here and be fed through the standard LessWrong algorithm.... | https://www.lesswrong.com/posts/Y8rEA4e4DxafmeAbW/the-inner-ring-by-c-s-lewis |
# Bayesian inference without priors
_Epistemic status: party trick_
Why remove the prior
--------------------
One famed feature of Bayesian inference is that it involves prior probability distributions. Given an exhaustive collection of mutually exclusive ways the world could be (hereafter called ‘hypotheses’), one ... | https://www.lesswrong.com/posts/bwjfXXf3GmGSe6rrP/bayesian-inference-without-priors |
# social lemon markets
> I refuse to join any club that would have me as a member.
— Groucho Marx
Alice and Carol are walking on the sidewalk in a large city, and end up together for a while.
> "Hi, I'm Alice! What's your name?"
Carol thinks:
> If Alice is trying to meet people this way, that means she doesn't ha... | https://www.lesswrong.com/posts/wfAF4v8kikxJoNqfL/social-lemon-markets |
# NIH Cancer Myths Myths
The NIH has a page called [Cancer Myths and Misconceptions](https://www.cancer.gov/about-cancer/causes-prevention/risk/myths) that you come across if you end up looking into cancer for long enough, aimed at bio-illiterate patients and their families.
Around half the things on that page are wr... | https://www.lesswrong.com/posts/ixJMcbizgmXRXxFbT/nih-cancer-myths-myths |
# The first future and the best future
It seems to me worth [trying to slow down AI development](https://worldspiritsockpuppet.substack.com/p/lets-think-about-slowing-down-ai) to steer successfully around the shoals of extinction and out to utopia.
But I was thinking lately: even if I didn’t think there was any chanc... | https://www.lesswrong.com/posts/Ejy4rRpGwsr9fCriP/the-first-future-and-the-best-future |
# "Why I Write" by George Orwell (1946)
People have been [posting](https://www.lesswrong.com/posts/2x8EJnEu4JZMiNbxd/this-is-water-by-david-foster-wallace#comments) great [essays](https://www.lesswrong.com/posts/Y8rEA4e4DxafmeAbW/the-inner-ring-by-c-s-lewis) so that they're "fed through the standard LessWrong algorith... | https://www.lesswrong.com/posts/Li4evP8nL7Xg4Wjjw/why-i-write-by-george-orwell-1946 |
# Improving Dictionary Learning with Gated Sparse Autoencoders
***Authors**: Senthooran Rajamanoharan*, Arthur Conmy*, Lewis Smith, Tom Lieberum, Vikrant Varma, János Kramár, Rohin Shah, Neel Nanda*
A new paper from the Google DeepMind mech interp team: Improving Dictionary Learning with Gated Sparse Autoencoders!
... | https://www.lesswrong.com/posts/vdoiWXeouGsZEYgrg/improving-dictionary-learning-with-gated-sparse-autoencoders |
# AXRP Episode 29 - Science of Deep Learning with Vikrant Varma
[YouTube link](https://youtu.be/4WNWeUQ7Hfc)
In 2022, it was announced that a fairly simple method can be used to extract the true beliefs of a language model on any given topic, without having to actually understand the topic at hand. Earlier, in 2021, ... | https://www.lesswrong.com/posts/PRZzA2wsFjaj4n8Pg/axrp-episode-29-science-of-deep-learning-with-vikrant-varma |
# Why I stopped being into basin broadness
There was a period where everyone was really into basin broadness for measuring neural network generalization. This mostly stopped being fashionable, but I'm not sure if there's enough written up on why it didn't do much, so I thought I should give my take for why I stopped f... | https://www.lesswrong.com/posts/utnS8tthytY8rT6fm/why-i-stopped-being-into-basin-broadness |
# Losing Faith In Contrarianism
Crosspost from [my blog.](https://benthams.substack.com/p/losing-faith-in-contrarianism)
If you spend a lot of time in the blogosphere, you’ll find a great deal of people expressing contrarian views. If you hang out in the circles that I do, you’ll probably have heard of [Yudkowsky]... | https://www.lesswrong.com/posts/qBDQQqQ9dWhJJ7Jt6/losing-faith-in-contrarianism |
# LLMs seem (relatively) safe
*Post for a somewhat more general audience than the modal LessWrong reader, but gets at my actual thoughts on the topic.*
In 2018 OpenAI [defeated the world champions](https://openai.com/research/openai-five-defeats-dota-2-world-champions) of Dota 2, a major esports game. This was hot on... | https://www.lesswrong.com/posts/ZxAWeiT8qNYppPbYA/llms-seem-relatively-safe |
# An Introduction to AI Sandbagging
**Summary:** Evaluations provide crucial information to determine the safety of AI systems which might be deployed or (further) developed. These development and deployment decisions have important safety consequences, and therefore they require trustworthy information. One reason wh... | https://www.lesswrong.com/posts/jsmNCj9QKcfdg8fJk/an-introduction-to-ai-sandbagging |
# Spatial attention as a “tell” for empathetic simulation?
(**UPDATE Sept 2024:** (1) **I still think “local spatial attention” is a thing as described below. But I no longer think that it functions as a ‘tell’ for empathetic simulation, as in the headline claim.** I have an alternate theory I like much better. Detail... | https://www.lesswrong.com/posts/7Pt9fogptmiSduXt9/spatial-attention-as-a-tell-for-empathetic-simulation |
# Scaling of AI training runs will slow down after GPT-5
*My credence: 33% confidence in the claim that the growth in the number of GPUs used for training SOTA AI will slow down significantly directly after GPT-5. It is not higher because of (1) decentralized training is possible, and (2) GPT-5 may be able to increase... | https://www.lesswrong.com/posts/xyL5kb8RBGLiupGLf/scaling-of-ai-training-runs-will-slow-down-after-gpt-5 |
# Fundamental Uncertainty: Chapter 8 - When does fundamental uncertainty matter?
*N.B. This is a chapter in a book about truth and knowledge. It is the first draft. I have since revised it. You can find the most up-to-date info/version on the* [*book's website*](https://www.fundamentaluncertainty.com/)*.*
In the prev... | https://www.lesswrong.com/posts/zZpofq8Fapf3JrF2v/fundamental-uncertainty-chapter-8-when-does-fundamental |
# Duct Tape security
*This is a linkpost for* [*On Duct Tape and Fence Posts*](https://outsidetheasylum.blog/duct-tape-and-fence-posts/)*.*
Eliezer writes about [fence post security](https://hpmor.com/chapter/115). When people think to themselves "in the current system, what's the weakest point?", and then dedicate t... | https://www.lesswrong.com/posts/CLS4FijfEFHzc5HEv/duct-tape-security |
# [Concept Dependency] Concept Dependency Posts
*This is a Concept Dependency Post. It may not be worth reading on its own, out of context. See the backlinks at the bottom to see which posts use this concept.*
*See the backlinks at the bottom of the post. Every post starting with [Concept Dependency] is a concept de... | https://www.lesswrong.com/posts/237cdeNLxAGWhSbDD/concept-dependency-concept-dependency-posts |
# We are headed into an extreme compute overhang
If we achieve AGI-level performance using an LLM-like approach, the training hardware will be capable of running ~1,000,000s concurrent instances of the model.
Definitions
===========
Although there is [some debate](https://www.lesswrong.com/posts/icR53xeAkeuzgzsWP/ta... | https://www.lesswrong.com/posts/cRFtWjqoNrKmgLbFw/we-are-headed-into-an-extreme-compute-overhang |
# On Not Pulling The Ladder Up Behind You
Epistemic Status: Musing and speculation, but I think there's a real thing here.
I.
--
When I was a kid, a friend of mine had a tree fort. If you've never seen such a fort, imagine a series of wooden boards secured to a tree, creating a platform about fifteen feet off the gr... | https://www.lesswrong.com/posts/k2kzawX5L3Z7aGbov/on-not-pulling-the-ladder-up-behind-you |
# D&D.Sci Long War: Defender of Data-mocracy
*This is an entry in the 'Dungeons & Data Science' series, a set of puzzles where players are given a dataset to analyze and an objective to pursue using information from that dataset. *
**STORY (skippable)**
=====================
You have the excellent fortune to live un... | https://www.lesswrong.com/posts/hveHKFcs3LZh2rJXh/d-and-d-sci-long-war-defender-of-data-mocracy |
# Superposition is not "just" neuron polysemanticity
**TL;DR:** *In this post, I distinguish between two related concepts in neural network interpretability: polysemanticity and superposition.* Neuron polysemanticity *is the observed phenomena that many neurons seem to fire (have large, positive activations) on multip... | https://www.lesswrong.com/posts/8EyCQKuWo6swZpagS/superposition-is-not-just-neuron-polysemanticity |
# Plausibility of Getting Early Warning Shots because AIs can't coordinate?
I don't know if this is a well known argument that has already been responded to. If it is, just delete the post.
An implicit assumption most people make when discussion takeover risk, is that any misaligned agent will stave off enacting take... | https://www.lesswrong.com/posts/Trnaa9skzuWBJsH7A/plausibility-of-getting-early-warning-shots-because-ais-can |
# Refusal in LLMs is mediated by a single direction
*This work was produced as part of Neel Nanda's stream in the *[*ML Alignment & Theory Scholars Program*](https://www.matsprogram.org) *\- Winter 2023-24 Cohort, with co-supervision from Wes Gurnee.*
*This post is a preview for our upcoming paper, which will provide... | https://www.lesswrong.com/posts/jGuXSZgv6qfdhMCuJ/refusal-in-llms-is-mediated-by-a-single-direction |
# Two Vernor Vinge Book Reviews
Vernor Vinge is a legendary and recently deceased sci-fi author. I’ve just finished listening to the first two books in the Zone of Thought trilogy. Both books are entertaining and culturally influential. The audio versions are high-quality.
A Deepness in the Sky is about two spacefari... | https://www.lesswrong.com/posts/tifiH2h5GtE3hogmQ/two-vernor-vinge-book-reviews |
# Link: Let's Think Dot by Dot: Hidden Computation in Transformer Language Models by Jacob Pfau, William Merrill & Samuel R. Bowman
One consideration that is pretty important for AI safety is understanding the extent to which a model's outputs are aligned with its chain of thought.
This [paper](https://arxiv.org/abs/... | https://www.lesswrong.com/posts/Mq5LqZtP54BC8fXte/link-let-s-think-dot-by-dot-hidden-computation-in |
# So What's Up With PUFAs Chemically?
*This is exploratory investigation of a new-ish hypothesis, it is not intended to be a comprehensive review of the field or even a a full investigation of the hypothesis*.
I've always been skeptical of the seed-oil theory of obesity. Perhaps this is bad rationality on my part, bu... | https://www.lesswrong.com/posts/TNHfhG2EWyGPLeEyd/so-what-s-up-with-pufas-chemically |
# Constructability: Plainly-coded AGIs may be feasible in the near future
Charbel-Raphaël Segerie and Épiphanie Gédéon contributed equally to this post.
Many thanks to Davidad, Gabriel Alfour, Jérémy Andréoletti, Lucie Philippon, Vladimir Ivanov, Alexandre Variengien, Angélina Gentaz, Simon Cosson, Léo Dana and Die... | https://www.lesswrong.com/posts/y9tnz27oLmtLxcrEF/constructability-plainly-coded-agis-may-be-feasible-in-the |
# Playing Northboro with Lily and Rick
This afternoon Lily, Rick, and I ("Dandelion") played our first dance together, which was also Lily's first dance. She's sat in with [Kingfisher](https://www.kingfisherband.com/) for a set or two many times, but this was her first time being booked and playing (almost) the whole ... | https://www.lesswrong.com/posts/pMaXQAT2EAPRdwn9d/playing-northboro-with-lily-and-rick |
# [Aspiration-based designs] 1. Informal introduction
**Sequence Summary.** This sequence documents research by [SatisfIA](https://pik-gane.github.io/satisfia/), an ongoing project on non-maximizing, aspiration-based designs for AI agents that fulfill goals specified by constraints ("aspirations") rather than maximiz... | https://www.lesswrong.com/posts/6BerZtxLQLgMSzA8n/aspiration-based-designs-1-informal-introduction |
# [Aspiration-based designs] 2. Formal framework, basic algorithm
**Summary.** In this post, we present the formal framework we adopt during the sequence, and the simplest form of the type of aspiration-based algorithms we study. We do this for a simple form of aspiration-type goals: making the expectation of some var... | https://www.lesswrong.com/posts/wtTz6hyP6hnX5NiuA/aspiration-based-designs-2-formal-framework-basic-algorithm |
# [Aspiration-based designs] 3. Performance and safety criteria, and aspiration intervals
**Summary.** In this post, we extend the basic algorithm by adding criteria for choosing the two candidate actions the algorithm mixes, and by generalizing the goal from making the expected Total equal a particular value to makin... | https://www.lesswrong.com/posts/yi4pqB6G73dcTnatq/aspiration-based-designs-3-performance-and-safety-criteria |
# [Aspiration-based designs] Outlook: dealing with complexity
**Summary.** This teaser post sketches our current ideas for dealing with more complex environments. It will ultimately be replaced by one or more longer posts describing these in more detail. Reach out if you would like to collaborate on these issues.
Mul... | https://www.lesswrong.com/posts/PbXwdFnSC26Q96FG3/aspiration-based-designs-outlook-dealing-with-complexity |
# Things I tell myself to be more agentic
I'm sharing this here since being agentic has become somewhat of a meme in EA circles. Despite its meme status, I find the idea very helpful and empowering, though difficult to put into action.
This is written as a personal self-affirmation — my hope is that by writing it up ... | https://www.lesswrong.com/posts/8JvowdXv47B3vsPaa/things-i-tell-myself-to-be-more-agentic |
# List your AI X-Risk cruxes!
[I'm posting this as a very informal community request in lieu of a more detailed writeup, because if I wait to do this in a much more careful fashion then it probably won't happen at all. If someone else wants to do a more careful version that would be great!]
By crux here I mean some u... | https://www.lesswrong.com/posts/wFmzoktuvf2WqhNNP/list-your-ai-x-risk-cruxes |
# Disentangling Competence and Intelligence
I struggle quite a bit with publishing, so today I am just going to publish a bunch of hopefully relatively refined drafts. Any comments and questions are welcome, I would be grateful to elaborate on any area where I appear unclear. If you like, you can also let me know some... | https://www.lesswrong.com/posts/r9CRDdmiKNjegisdu/disentangling-competence-and-intelligence |
# The Prop-room and Stage Cognitive Architecture
This is a post on a novel cognitive architecture I have been thinking about for a while now, first as a conceptual playground to concretise some of my agent foundation ideas, and lately as an idea for a project that approaches the Alignment Problem directly by concretis... | https://www.lesswrong.com/posts/AvPcE4vhFy25Za6vG/the-prop-room-and-stage-cognitive-architecture |
# Big-endian is better than little-endian
This is a response to the post [We Write Numbers Backward](https://www.lsusr.com/blog/we-write-numbers-backward.html), in which lsusr argues that little-endian numerical notation is better than big-endian.[^5rgaakohy3p] I believe this is wrong, and big-endian has a significant... | https://www.lesswrong.com/posts/DCbz8vtYekubPFqiM/big-endian-is-better-than-little-endian |
# Open-Source AI: A Regulatory Review
*Cross-posted on* [*the EA Forum*](https://forum.effectivealtruism.org/posts/vCpgHCFzS3xaio2Lm/open-source-ai-a-regulatory-review)*. This article is part of a series of ~10 posts comprising a **2024 State of the AI Regulatory Landscape Review**, conducted by the *[***Governance Re... | https://www.lesswrong.com/posts/vzGC4zh73dfcqnFgf/open-source-ai-a-regulatory-review |
# AISC9 has ended and there will be an AISC10
The 9th AI Safety Camp (AISC9) just ended, and as usual, it was a success!
[Follow this link](https://docs.google.com/presentation/d/1YpRUVijWS1iVqCvqMV4k_GphuCZsJVUc2yIPsRcFxXo/edit#slide=id.g2c69f1e9cd2_1_0) to find project summaries, links to their outputs, recordings... | https://www.lesswrong.com/posts/wB8KTFem8FsZ8u3Az/aisc9-has-ended-and-there-will-be-an-aisc10 |
# Ironing Out the Squiggles
### Adversarial Examples: A Problem
The apparent successes of the deep learning revolution conceal a dark underbelly. It may seem that we now know how to get computers to (say) [check whether a photo is of a bird](https://xkcd.com/1425/), but this façade of seemingly good performance is be... | https://www.lesswrong.com/posts/H7fkGinsv8SDxgiS2/ironing-out-the-squiggles |
# Towards a formalization of the agent structure problem
In [Clarifying the Agent-Like Structure Problem](https://www.lesswrong.com/posts/moi3cFY2wpeKGu9TT/clarifying-the-agent-like-structure-problem) (2022), [John Wentworth](https://www.lesswrong.com/users/johnswentworth) describes a hypothetical instance of what he ... | https://www.lesswrong.com/posts/oxsBpx9v3bgxraiPj/towards-a-formalization-of-the-agent-structure-problem |
# Towards Multimodal Interpretability: Learning Sparse Interpretable Features in Vision Transformers
Executive Summary
=================
In this post I present my results from training a Sparse Autoencoder (SAE) on a CLIP Vision Transformer (ViT) using the ImageNet-1k dataset. I have created an interactive web app, '... | https://www.lesswrong.com/posts/bCtbuWraqYTDtuARg/towards-multimodal-interpretability-learning-sparse-2 |
# LLMs could be as conscious as human emulations, potentially
Firstly, I'm assuming that high resolution human brain emulation that you can run on a computer is conscious in normal sense that we use in conversations. Like, it talks, has memories, makes new memories, have friends and hobbies and likes and dislikes and ... | https://www.lesswrong.com/posts/ASmcQYbhcyu5TuXz6/llms-could-be-as-conscious-as-human-emulations-potentially |
# Why I'm doing PauseAI
GPT-5 training is probably starting around now. It seems very unlikely that GPT-5 will cause the end of the world. But it’s hard to be sure. I would guess that GPT-5 is more likely to kill me than an asteroid, a supervolcano, a plane crash or a brain tumor. We can predict fairly well what the c... | https://www.lesswrong.com/posts/sfWPjmfZY4Q5qFC5o/why-i-m-doing-pauseai |
# Introducing AI Lab Watch
I'm launching [AI Lab Watch](https://ailabwatch.org). I collected actions for frontier AI labs to improve AI safety, then evaluated some frontier labs accordingly.
It's a coll... | https://www.lesswrong.com/posts/N2r9EayvsWJmLBZuF/introducing-ai-lab-watch |
# The Intentional Stance, LLMs Edition
*In memoriam of Daniel C. Dennett. *
**tl;dr:** I sketch out what it means to apply Dennett's *Intentional Stance* to LLMs. I argue that the intentional vocabulary is already ubiquitous in experimentation with these systems therefore what is missing is the theoretical framework ... | https://www.lesswrong.com/posts/zjGh93nzTTMkHL2uY/the-intentional-stance-llms-edition |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.