NeurIPS D&B track in a nutshell: (1) An LLM-generated benchmark dataset (2) used to test performance of LLMs (3) evaluated via LLM-as-a-judge

This is an issue on multiple levels, and authors using those "shortcuts"👀 are equally responsible for this unethical behaviour

Few days left for Call for Applications: Italian National PhD in Artificial Intelligence – AI & Society (2025-2028) 🎓 𝟒𝟒 𝐟𝐮𝐥𝐥𝐲 𝐟𝐮𝐧𝐝𝐞𝐝 𝟑-𝐲𝐞𝐚𝐫PhD positions 📷 𝗔𝗹𝗹 𝗶𝗻𝗳𝗼 at phd-ai-society.di.unipi.it/applications/ 🗓️ 𝐃𝐞𝐚𝐝𝐥𝐢𝐧𝐞: 18 July 2025, 13:00 (Italian time)

🧐What are the properties of self-explanations in GNNs? What can we expect from them? We investigate this in our #ICML25 paper. Come to have a chat at poster session 5, Thu 17 11 am. w. Sagar Malhotra andrea passerini Stefano Teso

Some aspects of AI discourse seem to come from a different planet, oblivious to basic realities on Earth. AI for science is one such area. In this new essay, Sayash Kapoor and I argue that visions of accelerating science through AI should be considered unserious if they don't confront

C'è stato almeno un attimo nella vita di tutti nel quale abbiamo pensato che Saponara fosse il miglior trequartista del mondo.

All week during rebuttals, I have started each day with the same reminder: stay humble, stay kind, don't let this turn me mean. When I was doing PhD, reviewers never felt this mean. There is a bright-eyed student sitting on the other side, and such reviews will destroy

È il 2031. Manganiello ha ordinato la bonifica della laguna di Orbetello. Il controllo Var è stato adottato al posto del Telepass. Un busto di Collina campeggia al Vittoriano. L'arbitro Maresca è il nuovo ambasciatore italiano in Arabia Saudita.

The new OpenAI paper “Why Language Models Hallucinate” is more like PR than research. The claim that hallucinations arise because training/evaluation reward guessing over abstaining is decades-old (reject option classifiers, selective prediction).

Very happy to share I have 2 papers accept at #NeurIPS2025 🏄‍♂️🥳 arxiv.org/abs/2502.11245 studies shotcuts and identifiability of concept-based models arxiv.org/abs/2506.03784 studies representational similarity of many models (e.g. LLMs & SSL) See you in San Diego 🏜️

🤔When Does Closeness in Distribution Imply Representational Similarity? Find it out --from an Identifiability perspective-- in our recent paper, published at NeurIPS Conference 🎉 arxiv.org/abs/2506.03784 1/6🧵

It’s unfortunate that a bug in OpenReview allowed reviewer and AC identities to be extracted. That said, this is a chance to reflect on our practices: We should write our reviews as if our names were attached to them and visible to authors and the broader community, even when

Today (11am-2pm, poster 1102) you can have a chat with Andrea Pugnana who is presenting a work on how to bridge concept-based models and learning to defer;

Physics brings in dreamers, people who are possibly there even against their own career self-interests, because they believe in something bigger. I wish there were more of this in AI. It was more like that in the “old days”, but those days are almost gone.

Mandatory to have a chat with Samuele and Emanuele.