←
Dylan Bowman753 posts
Dylan Bowman avatar

Dylan Bowman

@dylanbowmanSF

currently: pre-deployment team lead @ApolloResearch. formerly: waiting in the void for my flesh to be wrought

◷ Joined October 2021▧ 105 Photos and videos
Dylan Bowman avatar
Dylan Bowman@dylanbowmanSF·Nov 13

Not sure how to square this with the METR claim that models are not really autonomously leading research right now. Possibly all gains have just been perf eng related?

This slide from the Anthropic IPO roadshow was a big factor in my decision to leave. It’s an insane plot and I’m surprised they made it public. It shows the compute efficiency of pretraining over time. The y-axis is a **log scale**: they are seeing continued superexponential algorithmic progress. Mythos was not a one-time deviation, but a point on a superexponential. They could already retrain Mythos with a fraction of the compute originally needed. It doesn’t strictly imply a software-only intelligence explosion- they have been spending more on R+D. But algorithmic improvements do not require ridiculous creative genius and are on the verge of being automated. You do the math.

Post attachment
Dylan Bowman avatar
Dylan Bowman@dylanbowmanSF·Nov 13

Even Anthropic’s product team is rationalists I guess

Post attachment

Introducing Claude Meetings: add Claude to your team’s video call and it builds docs, whiteboards, and other artifacts with you while you’re still talking. Starting to roll out in beta on Team and Enterprise plans.

1:03
Neolithic avatar
⟳ Dylan Bowman retweeted
Neolithic@NeolithicAI·Nov 10

Introducing RIDArena A new control setting for evaluating rogue internal deployments and monitor sabotage. This is Neolithic's first AI safety infra release, built in collaboration with @redwood_ai

Post attachment
Dylan Bowman avatar
Dylan Bowman@dylanbowmanSF·Nov 9

Lmao (from Jared Kaplan)

Post attachment

Jared Kaplan on why Anthropic decided to be a public benefit corporation (instead of a nonprofit), in a deposition

Post attachmentPost attachmentPost attachmentPost attachment
Dylan Bowman avatar
Dylan Bowman@dylanbowmanSF·Nov 8

The report also glosses over the biggest plausible alternative to researcher uplift driving the acceleration, which is data uplift (like RL environment pipelines and automated quality control)

Dylan Bowman avatar
Dylan Bowman@dylanbowmanSF·Nov 8

Tiny tiny bit towards longer timelines because of this sentence: “Anthropic’s internal benchmarks are increasingly saturated”. This sentence is about AECI evals but maybe is also true of their production evals (low conf, hence “tiny tiny”)

Dylan Bowman avatar
Dylan Bowman@dylanbowmanSF·Nov 8

I have finished this report and learned approximately nothing (because of redactions) and updated a tiny tiny bit towards longer timelines.

Since August, a team within METR has worked with internal Anthropic information to assess whether AI itself is accelerating AI development. We are now publishing a report summarizing the results of this risk assessment.

Dylan Bowman avatar
Dylan Bowman@dylanbowmanSF·Nov 6

I don’t understand the point of this post. Most of the difficulty in continual learning is reward modeling but they just demo everything on chess puzzles…

Production traffic is full of real experience from economically valuable work, but very little of it currently makes it back into training. We built a recipe for production RL that steadily improves a model straight from its deployment’s data stream, then redeploys it to keep learning. We are sharing some of our initial learnings from a recipe capable of continual learning in production.

0:10
Dylan Bowman avatar
Dylan Bowman@dylanbowmanSF·Nov 4

This is pretty clearly so they can trigger OpenAI’s merge-and-assist clause and announce that they’re acquiring OpenAI the day of the IPO

Anthropic updated its S1 this morning. Notable additions re automated AI R&D and human oversight of it: "[...] our internal estimates and our expectations regarding future models, rather than our current models, indicate that we may imminently meet our Responsible Scaling Policy’s definition of 'dramatically accelerated progress attributable to AI.' We have not designated any of our current models as exceeding this threshold, and we continue to assess the underlying data. We are also evaluating the implications of meeting the AI research and development capability threshold and will release further reports in the future as we set forth in our Responsible Scaling Policy." "We are also considering additional measures, including updates to our Responsible Scaling Policy, to require human review and approval of certain decisions in our research and development processes as our models perform an increasing share of that work." "We believe our investments in safety techniques will enable us to manage these risks and responsibly realize substantially more leverage from Claude than we do today in developing AI models. However, we do not anticipate being able to safely use Claude to fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding, if and when that becomes technically possible. Accordingly, we are considering updates to our Responsible Scaling Policy to keep humans in control of frontier model development at Anthropic." sec.gov/Archives/edgar/data/…

Post attachmentPost attachmentPost attachment