Twitxy Saved

Ryan Greenblatt

Verified account This account is verified. Learn more

@RyanGreenblatt

MTS at @METR_Evals, focused on using investigations to make risk-relevant information about what's going on inside AI companies public.
lesswrong.com/users/ryan_gre… Joined September 2023
11 Following 25K Followers 2.3K Posts

Latest posts 5

Pinned post

Ryan Greenblatt Verified account @RyanGreenblatt
This report on AI R&D automation (RSI) is a big update towards ASI within 1-2 years. It says AI progress is sped up by RSI right now by around 2x. In AI 2027, this happens around March 2027.

I was seriously wrong about RSI speed up; I previously expected much less in 2026 (around 1.15x-1.3x). It seems as though there was specific privately available evidence that would have changed my mind.

I haven't seen an unredacted version of this report; this is purely about the public report. Because so much of the underlying evidence was redacted from the public report, I think conclusions about things other than the observed 2x faster rate of progress on benchmarks depend significantly on trust for the authors of the report. (E.g., conclusions about whether this speed up is from RSI depend on trust.) I do trust the authors of the report a lot to be careful and rigorous and generally do a good job. (Note that I also work at METR.) Less extensive redactions would greatly reduce this need for trust.

Since in my current role I may have access to other confidential information that could be relevant to timelines one way or another, I unfortunately cannot state my all-things-considered views on ASI timelines or the probability of a software intelligence explosion right now. Here, I’ve done some analysis of the implications using my previous public statements.

A 2x acceleration in overall AI progress from RSI corresponds to a much larger than 2x RSI speed up for AI software R&D (algorithms and data); I did an estimate using my public preexisting modeling of drivers of AI progress and the public report which yielded a very rough estimate of 4.5x for how much RSI is currently accelerating AI software progress (relative to no AI automation). In the AI 2027 scenario, this much acceleration happens around March 2027.

Here are the worked details of this estimate based on my public modeling. There are two main reasons RSI speed up of AI software R&D must be larger:

- Not all of AI progress is from better software (some is from scaling up compute in training runs). I've previously speculated that around 2/3 of AI progress is from software improvements which implies that 2x overall AI progress acceleration requires 3x faster software progress.
- Diminishing returns: A fixed amount of progress requires more work over time, so to ongoingly maintain 3x faster software progress via AI automation, you need increasingly large speed boosts from AI automation (as in, speed ups relative to how fast software R&D would go without AI automation). Based on this report, it seems that there is a substantial amount of additional AI progress from automation that has occurred.

We can estimate the effects of diminishing returns by first determining how many additional years of software progress we've seen (relative to the preexisting trend) and then estimating how much software R&D would slow down after a year of software progress if you held inputs fixed due to diminishing returns. I've previously estimated software R&D would slow down by around 2x for each year of software progress if you held inputs fixed (lesswrong.com/posts/FG54euEA…) and based on the report, it seems as though we've perhaps seen roughly 0.35-1.0 years of additional software progress incorporated into trained models so far (geometric mean of ~0.6). This combines to 3 * 2^(~0.6) = ~4.5.
(Relevant public modeling and numbers I'm using: lesswrong.com/posts/FG54euEA…, lesswrong.com/posts/Jga7PHMz…)
www.lesswrong.com ryan_greenblatt's Shortform — LessWrong Comment by ryan_greenblatt - I put roughly 50% probability on feasibility of software-only singularity.[1] (I'm probably going to be reinventing a bunch of the compute-centric takeoff model in...
METR @METR_Evals Nov 5
1. Benchmarks provide one piece of evidence for acceleration. Since Oct 2025, AECI scores for Anthropic’s models have grown at 1.9x [95%: 1.5–2.8] the earlier rate. Analysis choices affect this estimate, and it remains unclear how benchmark scores map to real-world capabilities.
View quoted post on X
56 replies 219 reposts 1.4K likes 316K views View on X
Ryan Greenblatt Verified account @RyanGreenblatt
The recent discussion around the AI R&D report supports the view that evidence transparency is very important.

I trust the authors and thus their estimate of large speed up from ai (~2x with very high uncertainty), but it's reasonable for people to want to see the key evidence!
Ajeya Cotra Verified account @ajeya_cotra Nov 5
I am sad that so much of the underlying evidence in this report was redacted. I think evidence transparency is very important for the world to figure this out in time. planned-obsolescence.org/p/evidence-abo…
View quoted post on X
8 replies 12 reposts 188 likes 9.5K views View on X
Ryan Greenblatt Verified account @RyanGreenblatt
For people following the weeds of the modeling, I made a small mistake in my math. After correcting this you get 3.8x instead of 4.5x for the software R&D multiplier from RSI. This still corresponds to March 2027 in AI 2027.

Sorry about the error.
Ryan Greenblatt Verified account @RyanGreenblatt Nov 6
Minor correction: I said "... around 2/3 of AI progress is from software improvements [so] 2x overall AI progress acceleration requires 3x faster software progress.", this has a small error, it should be 2.5x faster for an overall estimate of ~3.8x rather than 4.5x.

The error is that you still get progress from training compute, so the calculation to get how much faster software needs to be is

(2/3) x + 1/3 = 2

x = (2-1/3)/(2/3) = 2.5.

Sorry about the error!
View quoted post on X
6 replies 14 reposts 275 likes 10K views View on X
Ryan Greenblatt Verified account @RyanGreenblatt
Based on this article, OpenAI has made false accusations about me. The three researchers fired from OpenAI were told I facilitated architecture leaks to The Information. This is false.

A journalist from The Information did reach out to me shortly before that article came out. I did not comment at all beyond asking the journalist to make it clear that a blog post I wrote which was referenced in the article was preexisting and public.

The article also mentions an internal OpenAI monitorability proposal. I have heard that this memo contained highly sensitive architecture information, and that OpenAI accused me of accessing and reading it. This accusation is also false.

As reported in the article, I viewed a version of the proposal that did not have sensitive IP and was approved for sharing with me. I believe OpenAI now knows this accusation was based on a conflation of these two documents, though I do not know if they have clearly retracted that accusation.

I do not and have never solicited proprietary information, and the three employees who were recently fired from OpenAI never leaked IP to me. As the article mentions, I did have Signal chats with them that had disappearing messages on; they’re professional contacts that I’ve had on Signal for a while, and use of Signal is common in the tech industry. During the Hugging Face investigation, I was careful to avoid sending Signal messages with confidential information, though I did sometimes discuss high level things on Signal calls. As far as I understood, this was normal.
Max Zeff Verified account @ZeffMax Oct 23
New: The three researchers OpenAI fired last month spent the summer fighting to keep AI's chain-of-thought process visible.

Then, OpenAI asked to see their Signal chats. A week later, they were let go.

More below from @keachhagey, @berber_jin1, and me.

wsj.com/tech/ai/the-si…
View quoted post on X