Why hello there

New preprint!📢 There's been lots of exciting work recently on identifying gaps in and enhancing LLM reasoning. But it's equally important to take a step back and consider when doing so might HURT performance. Check out our paper for a framework!

CoT prompting can actually hurt LLM performance in some tasks. The paper shows LLMs and humans share similar limitations when forced to explain their thinking Identifies specific scenarios where asking LLMs to explain reduces their accuracy. i.e. When thinking out loud makes

Chain of thought can hurt LLM performance 🤖 Verbal (over)thinking can hurt human performance 😵‍💫 Are when/why they happen similar? Come find out at our poster at West-320 ⏰11am tomorrow! #ICML2025

One of the better posters I saw today at #icml25 This gets at the root of the problems we were thinking about when we conceived and wrote the CoT paper.

Shoutout to all the Princeton University researchers participating in ICML Conference #ICML2025 Browse through some of the cutting edge research from AI Lab students, post-docs and faculty being presented this year: pli.princeton.edu/blog/2025/prin…

Thanks so much for the excellent coverage and stopping by our poster at ICML! It was a pleasure to share our work with you!

Check out this sick video made by Ryan Liu about how LLMs can, counterintuitively, backfire when using chain-of-thought prompting!

Stay tuned 👀

How come LLM agents can carry out remarkable tasks like coding full-stack apps but still fall for poorly crafted pop-up scams? We formalize this using the psychological concept of motivational vigilance. Come to our PragLM spotlight talk (11:15 am 520B) and poster (1:30-2:30 pm

Everything online exists because someone had a reason to put it there. 🧐 LLMs process internet data, but do they consider why something was said 🤨 in the first place? NO 🛑 - in real online recommendations, LLMs get < .2 correlation with rationally weighing others' intent 🥴

New Anthropic paper tests whether LLMs notice motives behind messages and adjust trust accordingly. It asks whether they can tell when someone has a hidden motive, like being paid to promote something. Shows a weakness in how current LLMs judge trust. In simple test cases, the

We use LLMs for everyday tasks—research, writing, coding, decision-making. They remember our conversations, adapt to our needs and preferences. Naturally, we trust them more with repeated use. But this growing trust might be masking a hidden risk: what if their beliefs are