@pybeebee

CS PhD Candidate @Yale | prev. Research Intern @Google

pybeebee.com
Joined September 2015
  • Tweets115
  • Following187
  • Followers170
48Media
🎉Thrilled to share that RLMF is headed to #COLM2026! In this work we introduce 𝗥𝗟 𝘄𝗶𝘁𝗵 𝗺𝗲𝘁𝗮𝗰𝗼𝗴𝗻𝗶𝘁𝗶𝘃𝗲 𝗳𝗲𝗲𝗱𝗯𝗮𝗰𝗸 (𝗥𝗟𝗠𝗙) & metacognitive data selection. It outperforms standard RL while improving metacognitive monitoring in LLMs 🧠✨ Details below!👇
Gabrielle Kaili-May Liu (Kaili) ✈️ COLM@pybeebeeJul 1
🔥 New paper alert!! Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs 🔥 Can a model that learns to 𝘫𝘶𝘥𝘨𝘦 𝘪𝘵𝘴𝘦𝘭𝘧 get even better at tasks—and learn to be honest about what it doesn't know? Turns out: yes🤯 Details🧵👇
1
5
1
29
3,116
I will be at #COLM2026 presenting our paper “Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs”! 🧠⚙️ Come by to chat about LLM uncertainty, metacognition, post-training, and more! 📍Imp. Ballroom #23 🗓️Thu, Oct 8 | 4:40-6:30p
Gabrielle Kaili-May Liu (Kaili) ✈️ COLM@pybeebeeJul 8
🎉Thrilled to share that RLMF is headed to #COLM2026! In this work we introduce 𝗥𝗟 𝘄𝗶𝘁𝗵 𝗺𝗲𝘁𝗮𝗰𝗼𝗴𝗻𝗶𝘁𝗶𝘃𝗲 𝗳𝗲𝗲𝗱𝗯𝗮𝗰𝗸 (𝗥𝗟𝗠𝗙) & metacognitive data selection. It outperforms standard RL while improving metacognitive monitoring in LLMs 🧠✨ Details below!👇
2
19
481
🎈🎉 Delighted to announce that our paper has been accepted to #EMNLP2026 Main! This work studies the consistency & stability with which LLMs use linguistic markers (e.g., "probably", "likely") to convey their internal confidence. Check the original 🧵 for more details!👇
Gabrielle Kaili-May Liu (Kaili) ✈️ COLM@pybeebeeJun 1
🔥 Excited to share my new preprint: Can LLMs Use Linguistic Uncertainty Markers to Reliably Reflect Intrinsic Confidence? 🔥 When an LLM says "I think" or "probably," does it actually mean something consistent internally? The answer: not really 😬 Check out details in 🧵(1/n):
1
3
15
777
Metacognition is increasingly seen as a cornerstone of capable, trustworthy AI. Yet how to define, measure, or improve it in LLMs remains unclear. Our paper presents the first comprehensive overview of this emerging field!👇 📄 arxiv.org/pdf/2607.11881 🔗 github.com/yale-nlp/LLM-Meta…
5
1
44
2,334
Gabrielle Kaili-May Liu (Kaili) ✈️ COLM retweeted
Bryan Li@bryanlicsJul 17
Given a request, AI systems generate long-form reports. Report evaluation relies on nuggets: essential & granular facts. Prior work on nugget generation treats them as short statements. Our #ICTIR2026 oral proposes DoGMaTiQ, an automated pipeline for Q&A nuggets. 1/6🧵
1
4
9
864
Come to the AI4GOOD Workshop @ #ICML 2026 to learn more about how reasoning affects the faithfulness of models' verbalized uncertainty & what we can do about it! 📍Grand Ballroom 103 🗓️Fri, July 10 | 11:25-13:00 & 16:20-17:00 🔗arxiv.org/pdf/2606.03969
Gabrielle Kaili-May Liu (Kaili) ✈️ COLM@pybeebeeJun 3
Also glad to share this was accepted as a ✨spotlight✨to the Trustworthy AI4GOOD Workshop @ #ICML 2026!
1
7
554
Gabrielle Kaili-May Liu (Kaili) ✈️ COLM retweeted
Leyao Wang@LeyaoWang888Jul 6
🔥 New paper alert!! Time to REFLECT: Can We Trust LLM Judges for Evidence-based Research Agents? 🔥 We increasingly trust LLM judges to supervise deep research agents. But can we trust the judges themselves? Turns out: even frontier models score below 50% 😯 Details 🧵👇
1
3
8
571
🔥 New paper alert!! Reinforcement Learning with Metacognitive Feedback Elicits Faithful Uncertainty Expression in LLMs 🔥 Can a model that learns to 𝘫𝘶𝘥𝘨𝘦 𝘪𝘵𝘴𝘦𝘭𝘧 get even better at tasks—and learn to be honest about what it doesn't know? Turns out: yes🤯 Details🧵👇
4
19
5
64
12,000
more replies
(19/n) Check out our paper for more interesting insights and analysis, or if you're interested in RL, LLM metacognition, or calibration! Feedback and questions welcome :)) 📄 Paper: arxiv.org/pdf/2606.32032 🔗 Github: github.com/yale-nlp/RLMF
1
2
5
271
(20/n) Shoutout and thanks to my co-authors @clu_avi, @_galyo, Idan Szpektor, and @armancohan for helping make this happen!
1
1
198
Load more