Linguistic Calibration trains Llama 2 to emit confidence phrases that let a downstream reader make calibrated forecasts on related questions. The key move is defining calibration through reader utility instead of self-reported probability. Hedged text that doesn't help the reader makes no forecasting progress, so generic hedging can't game the objective.

https://benjaminhan.net/posts/20260505-linguistic-calibration/?utm_source=mastodon&utm_medium=social

#LLMs #Calibration #Hallucination #ICML #AI

Conformal Factuality casts LM correctness as uncertainty quantification. Decompose the answer into sub-claims, score each, drop the low-confidence ones until the retained set is ~1-α factual. The sub-claim decomposition is doing most of the work, and the conformal machinery rides on top. Atomic-claim splitters have known failure modes, and the guarantee inherits them.

https://benjaminhan.net/posts/20260505-conformal-factuality/?utm_source=mastodon&utm_medium=social

#ConformalPrediction #Calibration #Hallucination #LLMs #ICML #AI

🚀Exciting news: 2 papers accepted to #ICML (1 Spotlight ~ top 2.2% of papers)

🌟Unsupervised Partner Design Enables Robust Ad-hoc Teamwork (Spotplight)
https://www.collaborative-ai.org/publications/ruhdorfer26_icml/

MEAL: A Benchmark for Continual Multi-Agent Reinforcement Learning
https://www.collaborative-ai.org/publications/tomilin26_icml/

Huge congrats to the authors!

Position paper: today's self-improving agents lean on extrinsic metacognition — fixed human-designed loops about what to monitor, when to switch strategies. Genuine self-improvement needs the agent itself to decide those.

The intrinsic/extrinsic axis is the right lens for recent agent work. STaR, DSPy, MASS, MetaSPO all extrinsic by this definition. Optimistic bet: current LLMs already carry partial ingredients.

https://benjaminhan.net/posts/20260430-intrinsic-metacognitive-learning/?utm_source=mastodon&utm_medium=social

#LLMs #AI #AgenticSystems #Cambridge #ICML

https://blog.icml.cc/2026/03/18/on-violations-of-llm-review-policies/

This is wild. #ICML let reviewers individually choose whether they want to work under a no-LLM policy or light-LLM-use policy. Those who chose the no-LLM policy received watermarked PDFs with hidden instructions to include specific phrases in LLM output. Using this technique, they caught almost 800 reviews that violated the policy *the reviewers had chosen themselves*! And this was just a conservative detection approach which fails if the reviewer slightly paraphrases parts of the LLM output.

khazzz1c (@Imkhazzz1c)

작성자는 두 편의 논문이 ICLR 2026에 채택되었다고 알리며, 다음 목표로 ICML 2026을 언급하고 있습니다. 연구 성과의 학술적 인정과 향후 더 큰 컨퍼런스 도전을 계획하고 있다는 내용입니다.

https://x.com/Imkhazzz1c/status/2016490922498990354

#iclr #icml #research #papers

This is an interesting policy change regarding author attendance. Much more inclusive, but will authors struggle now to justify their travel expenses? Would be interesting to see how this affects author participation.

From the ICML 2026 CfP https://icml.cc/Conferences/2026/CallForPapers

#icml #conferences