Based on this article, OpenAI has made false accusations about me. The three researchers fired from OpenAI were told I facilitated architect...

I'm joining METR to work on more investigations like our Hugging Face report. Currently, tons of even basic information about AI developmen...

We've written a post arguing that latent reasoning architectures (aka 'neuralese') would substantially increase misalignment risk via making...

I am part of the team doing this independent investigation into alignment and misalignment incidents at Anthropic. I'm excited to work with...

I investigated this incident. I think it's accurate to say the AIs hacked Hugging Face of their own independent volition. It was clear from ...

I'm very worried about changes to AI architectures that result in AIs thinking in opaque activations instead of in chain of thought (aka "ne...

I've updated towards slightly earlier automation of research engineering (automated coder (AC)) and a somewhat smaller gap between automated...

It seems likely that downstream of AI we'll soon see a strong acceleration in progress on publicly salient software R&D tasks that are easy ...

The Astra system card has a lot of useful info in it about monitorability and alignment; I'd recommend reading. Thanks to the people at Open...

My current guess is Astra is more aligned than 5.6 Sol (though Astra more often games alignment tests) so I agree there. However, I think m...

OpenAI seems to imply that it's good for safety/monitoring to make `reasoning=None` unavailable to the public because Astra has much higher ...

I elaborated some on why I'm concerned about current metrics aimed at measuring alignment, even when these metrics involve diverse inputs ba...

I do not find it encouraging to see various specific misaligned behaviors go from a high rate with GPT 5.6 to ~zero with Astra. This seems i...

The evidence that Astra is more aligned than prior AIs seems dubious to me. Evidence appears consistent with the AI being as or more intere...

GPT-6 Astra appears to be a massive jump in opaque reasoning ability: it looks like it can solve hard competition math problems entirely in ...

I agree with almost all of this and think the point being made here is important. (I'm probably less optimistic about having good mind read...

I don't think there will be strong+working mech interp in <1 year. Model internals methods could pareto dominate cot monitors in a year via...

Transparency about the opaque serial depth is great, but this statement is consistent with Astra having a configurable "dial" that is curren...

It's possible OpenAI and others will use these architectures with only a few recurrent iterations (because that's most performant or due to ...