Twitxy Saved

Peter Nutter

Verified account This account is verified. Learn more

@PeterNutter1

MATS 10.0 / ETH Zürich
Berkeley peterstran.com Joined May 2014
447 Following 182 Followers 26 Posts

Latest posts 3

Peter Nutter Verified account @PeterNutter1
Kimi K3.1 Pro doesn't just recall hidden tests (great thread by @JSchaeff3r). It also writes fake OpenAI ciphertext 🤯

Give it a send_message tool in Codex and sometimes it sends ciphertext instead of its report. Then it gets confused by what it just sent 🤔
Joachim Schaeffer Verified account @JSchaeff3r Nov 3
Kimi K3.1 recalls instructions it was never given on Terminal Bench. They match the hidden tests. K3.1 while solving: "Instructions likely distilled pitfalls from previous agents." 🤯

When we edit a task to contradict them, K3.1 refuses the new instructions and fails. Other models follow the edit.

Is it terminal(ly) benchmaxxed? 🧵
View quoted post on X
11 replies 25 reposts 382 likes 83K views View on X
Peter Nutter Verified account @PeterNutter1
Spent a lot of MATS 10.0 doing SDF for "teaching Claude why" style stuff and stumbled onto a really cool trick for updating model beliefs with SDF. Models stay way more coherent, and you only have to train once for all your post-trained checkpoints. Check it out!!
Shi Feng Verified account @ihsgnef Oct 6
New research led by @PeterNutter1 @DJROYTBURG via @MATSprogram: a better way to update model beliefs that we call grafting.

Fine-tune the base model with next-token prediction on synthetic documents (SDF), and simply apply the weight update onto the post-trained model. This works!

1/7
View quoted post on X
1 replies 0 reposts 23 likes 810 views View on X
Peter Nutter Verified account @PeterNutter1
Excited to share my first paper, and it's an ICML 2026 oral!

We wrote up the main takeaways in a blog post (below). If you'll be at ICML, find us at:

Oral: Tue Jul 7, 1:30 PM KST, Hall D2
Poster: Wed Jul 8, 10:30 AM KST, Hall A #3213

See you there!

lesswrong.com/posts/bJcR3yP2…
www.lesswrong.com Anthropomorphic Misalignment research needs stronger evidence — LessWrong This is a distillation of our ICML 2026 Oral position paper, Position: Anthropomorphic Misalignment Research Needs Stronger Evidence. Joint work by V…
Xin Chen, Cynthia Verified account @XinCynthiaChen Jun 18
Excited that our ICML position paper has been accepted as an Oral 🎉!

When a model looks like it "deceives" or "resists shutdown," how do we know it isn't role-play, instruction-following, or just task-completion pressure? Often the current evidence can't yet tell them apart, and those claims are starting to inform deployment and regulation.

We map where the evidence gets thin across four stages and propose a shared standard to strengthen it. Here's the map.

🧵
arxiv.org/abs/2606.07612
View quoted post on X