iiinova — where your digital self keeps evolving

Reposted by akbir khan

Epoch AI @epochai.bsky.social · 08/05/2025

We’ve added four new benchmarks to the Epoch AI Benchmarking Hub: Aider Polyglot, WeirdML, Balrog, and Factorio Learning Environment! Before we only featured our own evaluation results, but this new data comes from trusted external leaderboards. And we've got more on the way 🧵

142

Reposted by akbir khan

Epoch AI @epochai.bsky.social · 08/05/2025

4. Factorio Learning Environment by Jack Hopkins, Märt Bakler , and @akbir.bsky.social This benchmark uses the factory-building game Factorio to test complex, long-term planning, with settings for lab-play (structured tasks) and open-play (unbounded growth). jackhopkins.github.io/factorio-lea...

jackhopkins.github.io

Factorio Learning Environment

Claude Sonnet 3.5 builds factories

121

Reposted by akbir khan

Johannes Gasteiger🔸 @gasteigerjo.bsky.social · 25/03/2025

New Anthropic blog post: Subtle sabotage in automated researchers. As AI systems increasingly assist with AI research, how do we ensure they're not subtly sabotaging that research? We show that malicious models can undermine ML research tasks in ways that are hard to detect.

143

akbir khan @akbir.bsky.social · 18/03/2025

control is a complimentary approach to alignment. its really sensible, practical and can be done now, even before systems are superintelligent. youtu.be/6Unxqr50Kqg?...

youtu.be

Controlling powerful AI

YouTube video by Anthropic

042

Reposted by akbir khan

Ethan Mollick @emollick.bsky.social · 25/02/2025

This is a crazy paper. Fine-tuning a big GPT-4o on a small amount of insecure code or even "bad numbers" (like 666) makes them misaligned in almost everything else. They are more likely to start offering misinformation, spouting anti-human values, and talk about admiring dictators. Why is unclear.

721443

Reposted by akbir khan

Hank Green @hankgreen.bsky.social · 28/01/2025

The fact that Deepseek R1 was released three days /before/ Stargate means these guys stood in front of Trump and said they needed half a trillion dollars while they knew R1 was open source and trained for $5M. Beautiful.

Trump announces 500B in AI funding. Five days ago.

396138081762

Reposted by akbir khan

Zack Witten @zswitten.bsky.social · 24/01/2025

Can anyone get a shorter DeepSeek R1 CoT than this?

3171

Reposted by akbir khan

Tom Everitt @tom4everitt.bsky.social · 23/01/2025

Process based supervision done right, and with pretty CIDs to illustrate :)

081

Reposted by akbir khan

Mark Riedl @markriedl.bsky.social · 21/01/2025

I don’t really have the energy for politics right now. So I will observe without comment: Executive Order 14110 was revoked (Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence)

29336

Reposted by akbir khan

Ethan Mollick @emollick.bsky.social · 15/01/2025

New randomized, controlled trial by the World Bank of students using GPT-4 as a tutor in Nigeria. Six weeks of after-school AI tutoring = 2 years of typical learning gains, outperforming 80% of other educational interventions. And it helped all students, especially girls who were initially behind.

1535388

Reposted by akbir khan

Ethan Mollick @emollick.bsky.social · 14/01/2025

Generative AI has flaws and biases, and there is a tendency for academics to fix on that (85% of equity LLM papers focus on harms)… …yet in many ways LLMs are uniquely powerful among new technologies for helping people equitably in education and healthcare. We need an urgent focus on how to do that

26911

Reposted by akbir khan

Ethan Mollick @emollick.bsky.social · 14/01/2025

On one hand, this paper finds adding inference-time compute (like o1 does) improves medical reasoning, which is an important finding suggesting a way to continue to improve AI performance in medicine On the other hand, scientific illustrations are apparently just anime now arxiv.org/pdf/2501.06458

2715

akbir khan @akbir.bsky.social · 13/01/2025

my metabolism is noticeably higher in london than the bay.

020

akbir khan @akbir.bsky.social · 10/01/2025

What can AI researchers do *today* that AI developers will find useful for ensuring the safety of future advanced AI systems? To ring in the new year, the Anthropic Alignment Science team is sharing some thoughts on research directions we think are important. alignment.anthropic.com/2025/recomme...

alignment.anthropic.com

Recommendations for Technical AI Safety Research Directions

2227

Reposted by akbir khan

Hank Green @hankgreen.bsky.social · 05/01/2025

My hottest take is that nothing makes any sense at all outside of the context of the constantly increasing value of human life, but that increase in value is so invisible (and exists in a world that was built for previous, lower values) that we constantly think the opposite has happened.

54175987

akbir khan @akbir.bsky.social · 04/01/2025

Nothing kills my excitement of returning to the US like the response i get from CBP officers.

170

Reposted by akbir khan

Andrew Lampinen @lampinen.bsky.social · 02/01/2025

Felix Hill was such an incredible mentor — and occasional cold water swimming partner — to me. He's a huge part of why I joined DeepMind and how I've come to approach research. Even a month later, it's still hard to believe he's gone.

Felix Hill and some other DMers and I after cold water swimming at Parliament Hill Lido a few years ago

712417

Reposted by akbir khan

Jane Wang @janexwang.bsky.social · 03/01/2025

A brilliant colleague and wonderful soul Felix Hill recently passed away. This was a shock and in an effort to sort some things out, I wrote them down. Maybe this will help someone else, but at the very least it helped me. Rest in peace, Felix, you will be missed. www.janexwang.com/blog/2025/1/...

janexwang.com

Felix — Jane X. Wang

From the moment I heard him give a talk, I knew I wanted to work with Felix . His ideas about generalization and situatedness made explicit thoughts that had been swirling around in my head, incohe...

26311

Reposted by akbir khan

Edward Grefenstette @egrefen.bsky.social · 30/12/2024

A few great papers out @ucldark.com this year. To single out two I love, there's the already well-cited paper on Debate by @akbir.bsky.social et al. which got best paper at ICML! [11/17]

141

akbir khan @akbir.bsky.social · 27/12/2024

just read this cover to cover in like 4 hours. strong recommend.

040

Reposted by akbir khan

Sam Bowman @sleepinyourhat.bsky.social · 18/12/2024

Alongside our paper, we also recorded a roundtable video featuring four of the paper’s authors discussing the results and their implications in detail:

youtube.com

Alignment faking in large language models

YouTube video by Anthropic

1212

Reposted by akbir khan

Sam Bowman @sleepinyourhat.bsky.social · 18/12/2024

New work from my team at Anthropic in collaboration with Redwood Research. I think this is plausibly the most important AGI safety result of the year. Cross-posting the thread below:

Title card: Alignment Faking in Large Language Models by Greenblatt et al.

512629

Reposted by akbir khan

Aengus Lynch @aengusl.bsky.social · 13/12/2024

NEW PAPER: Best-of-N Jailbreaking. We modify LLM inputs with simple, randomly generated augmentations and jailbreak frontier models across text, vision, and audio modalities. The algorithm is simple, scalable and highly effective.

151

akbir khan @akbir.bsky.social · 11/12/2024

if you can’t recognise o1s progress then you need scalable oversight

000

Reposted by akbir khan

SAKE.aM 🍶 @sakeandmiyazaki.bsky.social · 04/12/2024

Just finished the first two episodes of “Pantheon” on Netflix. I’m so moved. If you got free time and think you’ll love a Black Mirror-ish story themed adult animated series, PLEASEEEEE go watch this show and tell me what you think.

7648

Reposted by akbir khan

Max Roser @maxroser.bsky.social · 03/12/2024

Sometimes, the most important news is when something isn’t happening. In my new @OurWorldInData article, I highlight that US airlines have transported passengers for more than two light-years since the last plane crash. ourworldindata.org/us-airline-t...

ourworldindata.org

US airlines have transported passengers for more than two light-years since the last plane crash

Sometimes, the most important news is when something isn’t happening.

510023

Reposted by akbir khan

Stefan Schubert @stefanschubert.bsky.social · 03/12/2024

The Netherlands - a country with 26% of the UK's population - got almost as many consolidator grants as the UK this time. erc.europa.eu/news-events/...

081

Reposted by akbir khan

Sam Bowman @sleepinyourhat.bsky.social · 02/12/2024

If you're potentially interested in transitioning into AI safety research, come collaborate with my team at Anthropic! Funded fellows program for researchers new to the field here: alignment.anthropic.com/2024/anthrop...

alignment.anthropic.com

Introducing the Anthropic Fellows Program

37316

akbir khan @akbir.bsky.social · 02/12/2024

I’m recruiting Fellows to work with me on Aligning Superhuman models. alignment.anthropic.com/2024/anthrop...

330

Reposted by akbir khan

SAKE.aM 🍶 @sakeandmiyazaki.bsky.social · 22/11/2024

This nigga Kendrick went to therapy and got worse.

300140311686

akbir khan @akbir.bsky.social · 22/11/2024

The current structure provides you with a path where you end up with unilateral absolute control over the AGI. You stated that you don't want to control the final AGI but during this negotiation, you've shown to us that absolute control is extremely important to you www.lesswrong.com/posts/5jjk4C...

lesswrong.com

OpenAI Email Archives (from Musk v. Altman) — LessWrong

As part of the court case between Elon Musk and Sam Altman, a substantial number of emails between Elon, Sam Altman, Ilya Sutskever, and Greg Brockma…

020

akbir khan @akbir.bsky.social · 22/11/2024

impressed by the deepseek model

000

akbir khan @akbir.bsky.social · 22/11/2024

got into a weird habit of looking up someone’s thesis and reading the acknowledgments section to see who influenced them

010

akbir khan @akbir.bsky.social · 21/11/2024

incredibly cool work on demonstrating models truly do reason by Laura Ruis arxiv.org/abs/2411.12580

170