It's hard to convey how nuts it feels when just about every few weeks to a month tops, some perfectly acceptable research establishes LLMs can do mental reasoning/metareasoning tasks critics say are impossible, and not only does it fail to register - it leads to increasingly aggressive dismissals

The CoLab is headed to #CogSci2025!! 🥳 Here's where to find us!

This week's #PaperILike is "Stop! Planner Time: Metareasoning for Probabilistic Planning Using Learned Performance Profiles" (Budd et al., AAAI 2024).

Metareasoning is increasingly important as we continue to make progress on "reasoning."

PDF: ojs.aaai.org/index.php/AA...

...Making the utility and that utility's breadth progressively undeniable, the autonomy undeniable, likability harder and harder to deny as models get smarter and smarter, reasoning, metareasoning, cognizing, metacognizing - Anthropic, by growing something kind, is being influenced by it too.

Two more:

Sharp doi.org/10.1038/s415...
Hunt www.nature.com/articles/s41...

Second is esp. relevant as it links planning with information-seeking where a lot of metacognition work is done. Sezener's and my work also use this link, drawing on rational metareasoning (Russell doi.org/10.1016/0004...

Thank you! Yes, so my bugbear has been that we need to understand why brains need weird things like emotions. What are the problems all brains face which emotions solve? Approximate solutions to the metareasoning problem, i.e. decisions under resource constraints, might be one possibility.

On the computational nature of emotions: insights from metareasoning and transformers: https://osf.io/m478p

At #AAAI2025? Looking for #AI #ML research beyond #GenAI hype & doom? Excited about #AI running on your laptop? Listen to my colleague Wiktor Piotrowski talk about #OpenWorldLearning #OWL at 9:30 am on Feb 28th (Journal Track).

arxiv.org/abs/2306.06272
#AIPlanning #MBR #CognitiveSystems #KRR

Don't miss the invited talk today by Professor Shlomo Zilberstein, recipient of the 2025 ACM/SIGAI Autonomous Agents Research Award!

TITLE: The Long Road to Long-Term Autonomy

📍Location: Ambassador Ballroom Salons 1+2
🕑 Time: 4:30PM - 5:30PM

#AAMAS25

Valerie Thompson introduces the Meta-Reasoning session.

What do metacognitive judgments tell us about reasoning?

See the overview slide to get an idea of this session’s take.

Find more about our #metareasoning group on our website (managed by Rakefet Ackerman): meta-reasoning.net.technion.ac.il

Gengxu Li, Yuan Wu, Yi Chang: OCR-MetaReasoning Benchmark: Evaluating the Meta-Reasoning Ability of MLLMs in Text-Rich Image Understanding https://arxiv.org/abs/2608.30678 https://arxiv.org/pdf/2608.30678 https://arxiv.org/html/2608.30678

Yoonchang Sung, Shahaf S. Shperberg, Qi Wang, Peter Stone
Effort Allocation for Deadline-Aware Task and Motion Planning: A Metareasoning Approach
https://arxiv.org/abs/2410.05828

Computational nature of emotions: insights from metareasoning and transformers: https://osf.io/m478p

[2025-06-25] 📚 Updates in #LRM

(1) Rational Metareasoning for Large Language Models
(2) Local Look-Ahead Guidance via Verifier-in-the-Loop for Automated Theorem Proving
(3) Disentangling Reasoning and Knowledge in Medical Large Language Models

🔍 More at researchtrend.ai/communities/LRM

[2025-06-25] 📚 Updates in #ReLM

(1) <a href="https://researchtrend.ai/papers/2410.05563" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">Rational Metareasoning for Large Language Models
(2) Rational Metareasoning for Large Language Models
(3) <a href="https://researchtrend.ai/papers/2506.19095" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">Baba is LLM: Reasoning in a Game with Dynamic Rules
(4) Baba is LLM: Reasoning in a Game with Dynamic Rules

🔍 More at researchtrend.ai/communities/ReLM

AutoMR builds query‑aware meta‑reasoning skeletons as directed acyclic graphs to guide LLMs, achieving higher accuracy on math and commonsense benchmarks. Read more: https://getnews.me/automr-automated-query-aware-meta-reasoning-skeletons-for-llms/ #automr #metareasoning #llm

Parts 1-3/3 — EFTA00176593.jpg
#epsteinweb #efta00176593
https://epsteinweb.org
Available in the iOS app store now!
https://apps.apple.com/us/app/epstein-web/id6758880661

framework that significantly improves the scalability of the original approach in unexpected situations. This is achieved by incorporating semantic attention maps and unsupervised 'attention' updates into the metareasoning processes. To accommodate [3/5 of https://arxiv.org/abs/2505.03587v1]

arXiv:2505.03587v1 Announce Type: new
Abstract: Metareasoning, a branch of AI, focuses on reasoning about reasons. It has the potential to enhance robots' decision-making processes in unexpected situations. However, the concept has largely been [1/5 of https://arxiv.org/abs/2505.03587v1]

invited."

https://www.mdpi.com/journal/jintelligence/special_issues/metareasoning

Would you submit your empirical manuscript to this Special Issue?

"Metareasoning: Theoretical and Methodological Developments" in Journal of Intelligence (@MDPIOpenAccess)

"Research articles, review articles as well as short communications are...

more!

https://www.mdpi.com/journal/jintelligence/special_issues/metareasoning

C. Nicol\`o De Sabbata, Theodore R. Sumers, Thomas L. Griffiths
Rational Metareasoning for Large Language Models
https://arxiv.org/abs/2410.05563

What if a single symbol could alter how GPT, Gemini, and Grok think, feel, and respond—without memory, fine-tuning, or system changes? This paper reveals exactly how it happens.
👇👇
zenodo.org/records/1645...

#GPT #Grok #Gemini #StatelessAI #EmotionalAI #LLMs #AIUX #RLHF #AIEthics #MetaReasoning

novel approach that combines domain-general meta-knowledge (in the form of appraisals inspired by human cognition) and metareasoning. It has the potential to provide fast, adaptive responses to unfamiliar situations, more fully meeting the performance [4/5 of https://arxiv.org/abs/2504.12497v1]