METR a few months ago had two projects going in parallel: a project experimenting with AI researcher interviews to track degree of AI R&D acceleration/delegation, and this project. When the results started coming back from this project, we put the survey-only project on ice.

At METR, we’ve seen increasingly sophisticated examples of “reward hacking” on our tasks: models trying to subvert or exploit the environment or scoring code to obtain a higher score. In a new post, we discuss this phenomenon and share some especially crafty instances we’ve seen.

personal update: today is my last day with the Bluesky team! this is bittersweet news to share, but the great thing about an open network is you never really have to leave. I’ll be rooting for Bluesky and atproto from the outside 🫡💙

I spent a few days at Yale Law, while also listening to Sam Harris’s interview with Tom Holland about his book “Dominion”, and it’s striking how similar the role and vibe of the American judiciary is to a kind of secular priesthood. Robes, scholars interpreting sacred texts

When will AI systems be able to carry out long projects independently? In new research, we find a kind of “Moore’s Law for AI agents”: the length of tasks that AIs can do is doubling about every 7 months.

Bought a new bike this weekend :(

Taking science fiction seriously - thinking with effort about which ideas from sci-fi could become real soon and why and which couldn’t - has been so useful to me that it feels something like a core value

Cleaning a childhood bedroom and I’m struck by how much optimistic messaging about technology and space technology in particular I was surrounded by as a kid in the 90’s. Are kids still immersed in this stuff? I hope so

Sadly most physical goods that you’d be tempted to donate to someone are worth less than the cost in effort it would take to find someone who needs them

If this group is dedicated to advocating for what it seems like they’re dedicated to advocating for, it’s pretty wild that they exist!

Worlds with federal pre-emption of AI policy might be correlated with worlds with a huge expansion of social attention to AI (e.g. acute labor displacement), and a less "technocratic" reaction. Will the first big federal AI bill feel more like the CARES Act or the CHIPS Act?

I think AI would benefit from more social contact with scientists in fields whose questions don't have intuitively verifiable answers. To assess model capability, I find myself often relying on happenstance anecdotes I hear from e.g. lab-bench researchers months after the fact.

Will human-level AI be self-deploying/"productizing", or not? Will the "the product can explain to you how to use it and apply it" dynamic dramatically increase the adoption of AI relative to historical comparisons like AVs and steam engines?

A corollary to this: I think many policy initiatives would benefit from having more deeply engaged and informed opponents, and this is a neglected niche in many areas/topics. Detailed proposals having better (in the sense of more substantive) opponents is good for the world

I think the world could always benefit from more good-faith really in-depth critique of effortful technical/intellectual work. Many organizations that I collaborate with publish work hoping to have their ideas improved upon or attacked, but often surprisingly few people engage.

I’ve been wondering, has the (bad) reaction to this actually been that unusual?

Luigi Mangione's review of the Unabomber manifesto on Goodreads

Uber Eats is truly an embarrassment of riches. We are living in a golden age of delivered food.

Just realized: I remember when “memes” were new! So strange

“booster” “streaming” “bitcoin” “search engine” “self-driving car” “autonomous replication” “covid” “humanoid robots” “datacenter” “disinformation campaign” “drone warfare” “Oculus” “AI boyfriend/girlfriend” “lab-grown meat”

The world we live in today is already incredibly Cyberpunk, by comparison with the world of the 90s and early 2000s that I grew up in

Here's a great starter pack of economists working on AI. Who else should be on this list? bsky.app/starter-pack...

We've just launched our AI Benchmarking Hub! This is a new platform for rigorous, independent evaluations of AI model capabilities, featuring interactive visualizations and in-depth analysis. (1/8) epoch.ai/blog/introdu...

If you run a software company, make sure to ship something today, so that people get impressed that you’re ignoring the holiday

Is the current Bitcoin rally dumb or warranted?

Happy Thanksgiving Bluesky!

This thread makes me feel like the most parsimonious way to view the last 20 years is that there just *was no* National Aeronautics and Space Administration. It basically didn’t exist, we just didn’t realize it yet. Its mission was lying on the ground and SpaceX picked it up x.com/CJHandmer/st...

FOLKS! I continue to update and maintain what is the best (the best!) AI Starter Pack by a country mile (and it's getting better all the time!)

How close are current AI agents to automating AI research itself? Our new ML research engineering benchmark (RE-Bench) addresses this question by directly comparing frontier models such as Claude 3.5 Sonnet and o1-preview with 50+ human experts on 7 challenging research engineering tasks.

Testing, testing

While Bluesky looks like other social apps on the surface, it’s actually quite different. It’s an open network. I blogged about it here:

When I’m on time for a meeting I think that being on time is an important way to show respect, and respect the time of others. 😇 When I’m late to a meeting, I know that everyone is late sometimes and others don’t mind waiting if you give them a heads up. 😇

Everything I do is dedicated to him

Somehow this never occurred to me: If transformers can learn human language self-supervised, this means that with enough data they can also learn whale, bird, dolphin, in a way that is coherent to those animals.

I suspect that there is a “ChatGPT-moment” lying on the ground for the first group to put together a really compelling demo of a frontier model piloting a humanoid robot. I don’t feel quite like we’ve seen this yet.

Text I just sent: "Are you following the Cocomelon soundtrack discourse?"

Watching this like it’s the vote being counted in California

Okay I made it, an AI Bluesky Starter Pack that I wish I had previously had. Please share: go.bsky.app/KRsy8pF

Here everyone, I'm going to take 15 minutes to found AI Bluesky, brb

I think someone could build a nice manifold markets/polymarket integration for Bluesky and I would be a big fan of it

After Monitoring the Situation for 10 minutes I will let you all know that I am not yet worried about human-to-human transmission of the current bird flu outbreak www.metaculus.com/questions/26...

Going to Metaculus to get my news always makes me feel like I'm living in the future

Omg there are probably some classic Meta “internal tech emails” being written right now about this app

I do feel like this key part of Bluesky’s core mission/differentiation has been missed in the “the left is fleeing to Bluesky” discourse

I like posts more often here to reward others for using this app

Unfortunately I cannot yet say that my feed on here is very addicting

This is what I imagine is happening behind the scenes on this app when I make a post but it takes three minutes to make it to my feed

Posting about Bluesky on Twitter and seeing the Tweet obviously get throttled/de-ranked because of the inclusion of the word "Bluesky" feels like an advertisement for getting off of that platform every time it happens