LessWrong (Curated & Popular)

Summary: Superposition-based interpretations of neural network activation spaces are incomplete. The specific locations of feature vectors contain crucial structural information beyond superposition, as seen in circular arrangements of day-of-the-week features and in the rich structures. We don’t currently have good concepts for talking about this structure in feature geometry, but it is likely very important for model computation. An eventual understanding of feature geometry might look like a hodgepodge of case-specific explanations, or supplementing superposition with additional concepts, or plausibly an entirely new theory that supersedes superposition. To develop this understanding, it may be valuable to study toy models in depth and do theoretical or conceptual work in addition to studying frontier models.

Epistemic status: Decently confident that the ideas here are directionally correct. I’ve been thinking these thoughts for a while, and recently got round to writing them up at a high level. Lots of people (including [...]

The original text contained 5 footnotes which were omitted from this narration.

---

First published:
June 24th, 2024

Source:
https://www.lesswrong.com/posts/MFBTjb2qf3ziWmzz6/sae-feature-geometry-is-outside-the-superposition-hypothesis

---

Narrated by TYPE III AUDIO.

Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.This is a link post.TL;DR: We published a new paper on out-of-context reasoning in LLMs. We show that LLMs can infer latent information from training data and use this information for downstream tasks, without any in-context learning or CoT. For instance, we finetune GPT-3.5 on pairs (x,f(x)) for some unknown function f. We find that the LLM can (a) define f in Python, (b) invert f, (c) compose f with other functions, for simple functions such as x+14, x // 3, 1.75x, and 3x+2.

Paper authors: Johannes Treutlein*, Dami Choi*, Jan Betley, Sam Marks, Cem Anil, Roger Grosse, Owain Evans (*equal contribution)

Johannes, Dami, and Jan did this project as part of an Astra Fellowship with Owain Evans.

Below, we include the Abstract and Introduction from the paper, followed by some additional discussion of our AI safety [...]

---

First published:
June 21st, 2024

Source:
https://www.lesswrong.com/posts/5SKRHQEFr8wYQHYkx/connecting-the-dots-llms-can-infer-and-verbalize-latent

---

Narrated by TYPE III AUDIO.

This is a link post.I have canceled my OpenAI subscription in protest over OpenAI's lack ofethics.

In particular, I object to:

• threats to confiscate departing employees' equity unless thoseemployees signed a life-long non-disparagement contract
• Sam Altman's pattern of lying about important topics

I'm trying to hold AI companies to higher standards than I use fortypical companies, due to the risk that AI companies will exert unusualpower.

A boycott of OpenAI subscriptions seems unlikely to gain enoughattention to meaningfully influence OpenAI. Where I hope to make adifference is by discouraging competent researchers from joining OpenAIunless they clearly reform (e.g. by firing Altman). A few goodresearchers choosing not to work at OpenAI could make the differencebetween OpenAI being the leader in AI 5 years from now versus being,say, a distant 3rd place.

A [...]

---

First published:
June 18th, 2024

Source:
https://www.lesswrong.com/posts/sXhBCDLJPEjadwHBM/boycott-openai

---

Narrated by TYPE III AUDIO.

Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.This is a link post.New Anthropic model organisms research paper led by Carson Denison from the Alignment Stress-Testing Team demonstrating that large language models can generalize zero-shot from simple reward-hacks (sycophancy) to more complex reward tampering (subterfuge). Our results suggest that accidentally incentivizing simple reward-hacks such as sycophancy can have dramatic and very difficult to reverse consequences for how models generalize, up to and including generalization to editing their own reward functions and covering up their tracks when doing so.

Abstract:

In reinforcement learning, specification gaming occurs when AI systems learn undesired behaviors that are highly rewarded due to misspecified training goals. Specification gaming can range from simple behaviors like sycophancy to sophisticated and pernicious behaviors like reward-tampering, where a model directly modifies its own reward mechanism. However, these more pernicious behaviors may be too [...]

---

First published:
June 17th, 2024

Source:
https://www.lesswrong.com/posts/FSgGBjDiaCdWxNBhj/sycophancy-to-subterfuge-investigating-reward-tampering-in

---

Narrated by TYPE III AUDIO.

After living in a suburb for most of my life, when I moved to a major U.S. city the first thing I noticed was the feces. At first I assumed it was dog poop, but my naivety didn’t last long.

One day I saw a homeless man waddling towards me at a fast speed while holding his ass cheeks. He turned into an alley and took a shit. As I passed him, there was a moment where our eyes met. He sheepishly averted his gaze.

The next day I walked to the same place. There are a number of businesses on both sides of the street that probably all have bathrooms. I walked into each of them to investigate.

In a coffee shop, I saw a homeless woman ask the barista if she could use the bathroom. “Sorry, that bathroom is for customers only.” I waited five minutes and [...]

---

First published:
June 18th, 2024

Source:
https://www.lesswrong.com/posts/sCWe5RRvSHQMccd2Q/i-would-have-shit-in-that-alley-too

---

Narrated by TYPE III AUDIO.

ARC-AGI post

Getting 50% (SoTA) on ARC-AGI with GPT-4o

I recently got to 50%[1] accuracy on the public test set for ARC-AGI by having GPT-4o generate a huge number of Python implementations of the transformation rule (around 8,000 per problem) and then selecting among these implementations based on correctness of the Python programs on the examples (if this is confusing, go here)[2]. I use a variety of additional approaches and tweaks which overall substantially improve the performance of my method relative to just sampling 8,000 programs.

[This post is on a pretty different topic than the usual posts on our substack. So regular readers should be warned!]

The additional approaches and tweaks are:

• I use few-shot prompts which perform meticulous step-by-step reasoning.
• I have GPT-4o try to revise some of the implementations after seeing what they actually output on the provided examples.
• I do some feature engineering [...]

The original text contained 15 footnotes which were omitted from this narration.

---

First published:
June 17th, 2024

Source:
https://www.lesswrong.com/posts/Rdwui3wHxCeKb7feK/getting-50-sota-on-arc-agi-with-gpt-4o

---

Narrated by TYPE III AUDIO.

Have you heard this before? In clinical trials, medicines have to be compared to a placebo to separate the effect of the medicine from the psychological effect of taking the drug. The patient's belief in the power of the medicine has a strong effect on its own. In fact, for some drugs such as antidepressants, the psychological effect of taking a pill is larger than the effect of the drug. It may even be worth it to give a patient an ineffective medicine just to benefit from the placebo effect. This is the conventional wisdom that I took for granted until recently.

I no longer believe any of it, and the short answer as to why is that big meta-analysis on the placebo effect. That meta-analysis collected all the studies they could find that did "direct" measurements of the placebo effect. In addition to a placebo group that could [...]

---

First published:
June 10th, 2024

Source:
https://www.lesswrong.com/posts/kpd83h5XHgWCxnv3h/why-i-don-t-believe-in-the-placebo-effect

---

Narrated by TYPE III AUDIO.

Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.As an AI researcher who wants to do technical work that helps humanity, there is a strong drive to find a research area that is definitely helpful somehow, so that you don’t have to worry about how your work will be applied, and thus you don’t have to worry about things like corporate ethics or geopolitics to make sure your work benefits humanity.

Unfortunately, no such field exists. In particular, technical AI alignment is not such a field, and technical AI safety is not such a field. It absolutely matters where ideas land and how they are applied, and when the existence of the entire human race is at stake, that's no exception.

If that's obvious to you, this post is mostly just a collection of arguments for something you probably already realize. But if you somehow [...]

---

First published:
June 14th, 2024

Source:
https://www.lesswrong.com/posts/F2voF4pr3BfejJawL/safety-isn-t-safety-without-a-social-model-or-dispelling-the

---

Narrated by TYPE III AUDIO.

Preamble: Delta vs Crux

This section is redundant if you already read My AI Model Delta Compared To Yudkowsky.

I don’t natively think in terms of cruxes. But there's a similar concept which is more natural for me, which I’ll call a delta.

Imagine that you and I each model the world (or some part of it) as implementing some program. Very oversimplified example: if I learn that e.g. it's cloudy today, that means the “weather” variable in my program at a particular time[1] takes on the value “cloudy”. Now, suppose your program and my program are exactly the same, except that somewhere in there I think a certain parameter has value 5 and you think it has value 0.3. Even though our programs differ in only that one little spot, we might still expect very different values of lots of variables during execution - in other words, we [...]

---

First published:
June 12th, 2024

Source:
https://www.lesswrong.com/posts/7fJRPB6CF6uPKMLWi/my-ai-model-delta-compared-to-christiano

---

Narrated by TYPE III AUDIO.

Preamble: Delta vs Crux

I don’t natively think in terms of cruxes. But there's a similar concept which is more natural for me, which I’ll call a delta.

Imagine that you and I each model the world (or some part of it) as implementing some program. Very oversimplified example: if I learn that e.g. it's cloudy today, that means the “weather” variable in my program at a particular time[1] takes on the value “cloudy”. Now, suppose your program and my program are exactly the same, except that somewhere in there I think a certain parameter has value 5 and you think it has value 0.3. Even though our programs differ in only that one little spot, we might still expect very different values of lots of variables during execution - in other words, we might have very different beliefs about lots of stuff in the world.

If your model [...]

The original text contained 1 footnote which was omitted from this narration.

---

First published:
June 10th, 2024

Source:
https://www.lesswrong.com/posts/q8uNoJBgcpAe3bSBp/my-ai-model-delta-compared-to-yudkowsky

---

Narrated by TYPE III AUDIO.