akbir khan
@akbir.bsky.social
dumbest overseer at @anthropic
https://www.akbir.dev
https://www.akbir.dev
Reposted by akbir khan
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 🧵
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 🧵
May 8, 2025 at 3:00 PM
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 🧵
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 🧵
Reposted by akbir khan
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...
@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...
Factorio Learning Environment
Claude Sonnet 3.5 builds factories
jackhopkins.github.io
May 8, 2025 at 3:00 PM
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...
@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...
Reposted by akbir khan
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.
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.
March 25, 2025 at 4:03 PM
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.
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.
control is a complimentary approach to alignment.
its really sensible, practical and can be done now, even before systems are superintelligent.
youtu.be/6Unxqr50Kqg?...
its really sensible, practical and can be done now, even before systems are superintelligent.
youtu.be/6Unxqr50Kqg?...
Controlling powerful AI
YouTube video by Anthropic
youtu.be
March 18, 2025 at 3:22 PM
control is a complimentary approach to alignment.
its really sensible, practical and can be done now, even before systems are superintelligent.
youtu.be/6Unxqr50Kqg?...
its really sensible, practical and can be done now, even before systems are superintelligent.
youtu.be/6Unxqr50Kqg?...
Reposted by akbir khan
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.
February 25, 2025 at 9:01 PM
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.
This is the entire goal
It’s weird to live in a world where AI models are more aligned than the CEOs of the companies creating them
February 1, 2025 at 2:13 AM
This is the entire goal
Reposted by akbir khan
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.
Beautiful.
January 28, 2025 at 3:02 AM
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.
Beautiful.
Reposted by akbir khan
Can anyone get a shorter DeepSeek R1 CoT than this?
January 24, 2025 at 6:11 AM
Can anyone get a shorter DeepSeek R1 CoT than this?
Reposted by akbir khan
Process based supervision done right, and with pretty CIDs to illustrate :)
January 23, 2025 at 8:33 PM
Process based supervision done right, and with pretty CIDs to illustrate :)
Reposted by akbir khan
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)
Executive Order 14110 was revoked (Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence)
January 21, 2025 at 12:34 AM
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)
Executive Order 14110 was revoked (Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence)
R1 model is impressive
January 21, 2025 at 10:21 PM
R1 model is impressive
Reposted by akbir khan
Reposted by akbir khan
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.
And it helped all students, especially girls who were initially behind.
January 15, 2025 at 8:58 PM
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.
And it helped all students, especially girls who were initially behind.
Reposted by akbir khan
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
…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
January 14, 2025 at 5:45 PM
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
…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
Reposted by akbir khan
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
On the other hand, scientific illustrations are apparently just anime now arxiv.org/pdf/2501.06458
January 14, 2025 at 5:56 AM
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
On the other hand, scientific illustrations are apparently just anime now arxiv.org/pdf/2501.06458
my metabolism is noticeably higher in london than the bay.
January 13, 2025 at 3:49 PM
my metabolism is noticeably higher in london than the bay.
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/2025/recomme...
Recommendations for Technical AI Safety Research Directions
alignment.anthropic.com
January 10, 2025 at 9:03 PM
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/2025/recomme...
Reposted by akbir khan
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.
January 5, 2025 at 7:08 PM
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.
Reposted by akbir khan
Nothing kills my excitement of returning to the US like the response i get from CBP officers.
January 4, 2025 at 4:13 AM
Nothing kills my excitement of returning to the US like the response i get from CBP officers.
Reposted by akbir khan
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.
January 2, 2025 at 7:01 PM
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.