Xinyue Liu

🙌 Excited to share our new paper and my first project in my PhD journey! We show finetuning on a writing task unlocks verbatim recall of copyrighted books from authors not in the finetuning data. It’s been an incredible experience working with such an amazing group of people ✨

Today, we are introducing Inkling. Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available. thinkingmachines.ai/news/introduci… Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵

New Anthropic research: Agentic misalignment in Summer 2026. A year after our blackmail experiments, we found four more ways that today’s autonomous AI agents misbehave in simulations. Read more: alignment.anthropic.com/2026/agentic-m…

Unsurprisingly, things have gotten worse over time - check out the post for more, and thanks to @pangram for supplying the credits that funded this work Note that this was all post-hoc analysis, we didn't use Pangram when making decisions lesswrong.com/posts/r7FBQ8XD…

The context can be leaked!! How much information from context could your LLM be leaking? Say you are a healthcare practitioner using LLMs with a sensitive dataset to help users make a diagnosis. An attacker could come and retrieve this sensitive information using a prompt injection attack, which would be catastrophic. But what if we add a layer of defense? Would this still be the case even if we prompted the LLM not to output this information? What if we even add an LLM to detect if this information is being leaked? What about even stronger defenses that utilize differential privacy? Our #COLM 2026 paper addresses these questions: 📄 arxiv.org/abs/2512.16059

@icmlconf 🙏 I have made annotated slides available here: cs.princeton.edu/~arvindn/talks…

At humans&, we train models from the long-term impacts of their interactions with people. This requires prioritizing long-horizon multi-agent RL. We've developed and are excited to share an open-source, hardware-native 4-bit RL recipe, significantly accelerating training

@TuhinChakr @COLM_conf LFG🐐

Now accepted at @COLM_conf 🌁🦙Congratulations to my PhD student @irisiris_l on her first paper 🥂

A great lesson learned from #ACL2026 test of time award winner: sometimes it just takes time for people to appreciate your work

Lately I have been thinking a lot about AI for education (inspired by the amazing @alexisjross), and have been realizing that training models for education (both AI tutors and AI students) is non-trivial and bottle necked by user data, and how theory of mind plays a big role. I gave a tutorial on this at the Workshop on Innovative Use of NLP for Building Educational Applications @aclmeeting (virtually), where I talked about the privacy risks posed here: consent, data retention, memorization and context for personalization. slides⬇️, recording coming soon!

As my time in Pittsburgh comes to an end... I'm excited to share that I will be joining UIUC as an Assistant Professor in fall 2027! I will recruit PhD students in both ECE and CS. I’m looking for students and postdocs who are excited to continue building towards collaborative AI systems that learn and adapt through interaction.

Nice !!

unfortunately im gonna miss this but if ur there come and attend Niloofar’s talk!

Does LLM really need to be a helpful assistant all the time? No. If you want to simulate people, “perfectly helpful” could be the wrong objective. Meet OdysSim, a journey toward LLMs beyond assistants, as behavioral foundation models (10B tokens of real human behavior; 23 sim benchmarks, finally in one place. new open models: outperform or on par with GPT-5.5, Gemini 3.1, or Claude Opus 4.7 in many behavior-sim dimensions). Human behavior simulation is becoming essential. Agent evaluation needs realistic users before real users show up. Medical and classroom training need realistic patients and students. Social science needs synthetic participants at scale. But real people are not ideal assistants. Real patients panic or ignore good advice. Real students misunderstand. Real customers are vague, picky, impatient, or simply leave. Human behavior is messy, diverse, and often imperfect. Frontier LLMs are getting better at math, code, and long-horizon tasks. They are NOT getting better at simulating human behavior. If anything, they drift the other way: more assistant-ish, more homogeneous, fewer of the errors and quirks real humans show. This is no accident. The whole pipeline is built for helpfulness and task success, not behavioral realism. And you can't prompt your way out of that. So we rethink the recipe from scratch and release: 🧠 The OdysSim corpus: 21.4M real human interactions (~10B tokens) from 62 sources, every conversation retrofitted with social grounding (who is talking, and why) 📏 SOUL-Index: 23 human-behavior benchmarks unified into one suite across 5 axes 🤖 OSim-8B: open weights; tops more SOUL-Index benchmarks than any frontier model, acts more like a real user than any of them on τ-bench (nearly matching real humans in the reaction dimension), and writes far more human-like text along the way.

In film, "we'll fix it in post" is what you say when something went wrong on set and you don't want to redo it. AI research has made it our entire methodology: train the model, then patch whatever comes out. Our new ICML oral argues this can't be the basis of a science of AI. 🧵

How would you design a pretrained LLM that preserves output diversity AUTOMATICALLY after finetuning? Our method: learn a diverse “annotation” distribution from the pretraining data that conditions the generations, and then **don’t touch it when fine-tuning**! 1/

(1/n)🚨 Preprint alert🔥: Finetune an LLM on insecure code, and it may start recommending dangerous actions on totally unrelated prompts. That’s emergent misalignment (EM) We ask: can we catch EM inside the model during training?🧐 arxiv.org/abs/2606.07631 #AIsafety #llm #ai #ml

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Its capabilities exceed those of any model we’ve ever made generally available.

From op-eds in newspapers to NeurIPS position papers, AI is increasingly shaping long-form public discourse. Its arguments seem plausible, but beneath surface fluency, we find argument collapse: different LLMs converge to the same main & supporting arguments and structure.