dmm | Entries tagged with ai safety

Lean theorem prover seems to be the most convenient and practical (good enough for mathematicians to use it in real work, I am seeing plenty of such stories lately; there is a bit of controversy on whether to use Lean 3 or Lean 4).

The best quote according to "leanprover" twitter:

>Our favorite quote from the paper: "Monadic syntax is excellent for expressing stochastic algorithms, and working over finitely supported distributions avoids the need for integrability side conditions during proofs."

Links are in the comments.

"A second difficulty in communicating alignment ideas was based on differing ontologies. A surface-level explanation is that Japan is quite techno-optimistic compared to the west, and has strong intuitions that AI will operate harmoniously with humans. A more nuanced explanation is that Buddhist- and Shinto-inspired axioms in Japanese thinking lead to the conclusion that superintelligence will be conscious and aligned by default. One senior researcher from RIKEN noted during the conference that “it is obviously impossible to control a superintelligence, but living alongside one seems possible.” Some visible consequences of this are that machine consciousness research in Japan is taken quite seriously, whereas in the West there is little discussion of it."

***

I think it's time for us to start asking if, for example, GPT-4-produced simulations have associated subjective experience.

We have a feed-forward transducer in an autoregressive mode; each time a new token is produced by the feed-forward Transformer, the whole dialog including the just produced token is fed again to the input of the model, so there is a recurrent dynamics here (cf. section 3.4 of "Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention", arxiv.org/abs/2006.16236).

So I would not be too surprised if that process actually "feels what it says".

I am reading more and more LessWrong in recent months (mostly, after the Simulator theory by Janus (work done while at Conjecture) has been posted there in September).

I still think the Simulator theory is probably the single most important research breakthrough of 2022.

These days LessWrong is dominated by writing related to AI safety (the topic is made particularly acute by the recent progress in LLMs: ChatGPT and even more capable Bing Chat; no consensus whatsoever, of course, but I do think that GPT-3 release in May 2020 is, in some sense, an equivalent of the nuclear fission discovery on 19 December 1938, and that ChatGPT performance (+ Bing Chat clearly drastically enhanced capabilities even compared to that) is, in the same sense, an equivalent of the first working nuclear reactor on 2 December 1942, if one goes by "AI today is what nuclear energy has been back then" analogy).

So, one thing which might be useful is that there is GreaterWrong alternative viewer (which looks different from LessWrong default viewer and which can be visually tuned in terms of presentation style; also different default front page for the site if one uses GreaterWrong). Which viewer is better might depend on your device (display, browser, etc).

Another thing, Conjecture people tend to produce some of the best, most interesting articles there.

I'll put a few links into the comments.

www.lesswrong.com/posts/r2yTwkGt3kbQG2mXi/axrp-episode-19-mechanistic-interpretability-with-neel-nanda

To read my https://twitter.com/home more regularly (that's absolutely the best source of info at the moment).

A small fraction of today's catch:

New work by Janus

A new involved take on AI safety/alignment

(What's the right way to organize all that information?)

Links are in the comments (I think the new work by Janus is more important even for alignment, and is just overall more important of the two topics of this post)...

AI-safety-wise, the write-up, My AI Safety Lecture for UT Effective Altruism by Scott Aaronson is very nice reasonably objective and theory-friendly overview of the current state of AI safety as a field of science.

AI-progress-wise, ChatGPT based on roughly speaking GPT-3.5 has been released recently, with people doing tons of interesting things with it, including meaningful writing and software generation... This seems to be another major step-up.

This week, Nov 17-18, Thu-Fri, 8am-11:45am Boston time, "Quantum physics and the first-person perspective": www.essentiafoundation.org/quantum-physics-and-the-first-person-perspective/seeing/

JuliaCon 2023, juliacon.org/2023/ the call for proposals is posted, deadline Dec 18: pretalx.com/juliacon2023/cfp

I've spent more quality time focusing of two breakthroughs in understanding the nature and the behavior of machine learning models which came from the "penumbra" of "prosaic alignment" start-ups and which I wrote about in my previous two posts.

"Grokking is (more or less) solved." I took brief notes between Oct 21 and Oct 23: github.com/anhinga/2022-notes/tree/main/Grokking-is-solved

"Generative autoregressive models are similators." I took extensive notes between Oct 5 and Oct 23: github.com/anhinga/2022-notes/tree/main/Generative-autoregressive-models-are-similators

I am continuing to develop thoughts related to these topics, I am going to gradually write more about those topics in the comments.

The most interesting conceptual AI advances seem lately to come from "prosaic alignment" start-ups. These are companies which believe that the current trend of improving Transformer models is likely to lead straight to AGI, and that better understanding of the nature and properties of these model is key to AI safety (and, of course, it's also key to better AI capabilities).

And it is often the case that the key elements of work are done by people "on the edge", "in the penumbra" of those alignment start-ups.

In the previous post I mentioned the key new understanding of large Transformer models as simulators. That work has been done "while at Conjecture", but is not listed as directly coming from Conjecture (one of those "prosaic alignment" start-ups). I think the key people involved are still at Conjecture, but they seem to be trying to keep some distance between Conjecture and this work. I am continuing to take notes of those materials and commit them to GitHub (see links in the comments to the previous post).

Here is another one of those stories. Grokking is a phenomenon, where small Transformers look at a part of a mathematical structure for quite a while, and then rather suddenly transition to understanding the whole of that mathematical structure including the part they never see in training. It has been discovered in 2021 and has been a subject of a number of follow-up attempts to understand it.

The recent breakthrough has been done in mid-August by Neel Nanda who left Anthropic (perhaps the most famous of the "prosaic alignment" start-ups) a few months ago. And it looks like he has more or less solved the mysteries behind this phenomenon. I am going to continue studying his writings more. The links are in the comments.

Вот, наконец, кажется возник правильный подход к пониманию природы моделей вроде GPT-3 и разнообразного волшебства, с этим связанного:

www.lesswrong.com/posts/vJFdjigzmcXMhNTsx/simulators

Он говорит, что надо перестать думать про эти модели в терминах более старых AI-систем.

openai.com/blog/our-approach-to-alignment-research/

"Our approach to aligning AGI is empirical and iterative. We are improving our AI systems’ ability to learn from human feedback and to assist humans at evaluating AI. Our goal is to build a sufficiently aligned AI system that can help us solve all other alignment problems."

astralcodexten.substack.com/p/updated-look-at-long-term-ai-risks

The main takeaway is that no scenario is considered as much more likely than others by the best experts, and they all look more or less equally likely except for the "scenario not listed here" (which is rated as somewhat more likely than the listed scenarios).

Also people seem to be very optimistic for some reason (perhaps, they secretly believe in a benevolent G-d or benevolent aliens keeping an eye of us; otherwise their optimism is difficult to explain).

Scott Alexander summarizes the takeaways interesting for him as follows:

======= QUOTE =======

1. Even people working in the field of aligning AIs mostly assign “low” probability (~10%) that unaligned AI will result in human extinction

2. While some people are still concerned about the superintelligence scenario, concerns have diversified a lot over the past few years

3. People working in the field don't have a specific unified picture of what will go wrong