Excited to share our paper: "Chain-of-Thought Is Not Explainability"! We unpack a critical misconception in AI: models explaining their Chain-of-Thought (CoT) steps aren't necessarily revealing their true reasoning. Spoiler: transparency of CoT can be an illusion. (1/9) 🧵

News from CIIT Lab: the paper "RAGGAE for HERBS: Testing the Explanatory Performance of Ontology-powered #LLMs for Human Explanation of Robotic Behaviors" by agnese augello, E. Datteri, Antonio Lieto @[email protected], M. Rausa & N. Zagni has been accepted ICSR2025. Link: ciitlab.org/RAGGAE4HERBS_I…

🗓️ Oggi 6 luglio alle 18 su Radio Radicale🎙️ una nuova puntata di Puntini sull’AI con #AntonioLieto, Professore di Informatica all’Università di Salerno, Direttore CIIT Lab, su #IA e #Astrazione. Link ⬇️

🗓️ Oggi alle ore 18 su Radio Radicale 🎙️ una puntata di "Puntini sull’AI" dove parlerò di #AI, #Astrazione, #Ragionamento e #Metafore menzionando il lavoro su METCL (METAPHOR ELABORATION IN TYPICALITY-BASED COMPOSITIONAL LOGIC) accettato a #IJCAI2025: ciitlab.org/IJCAI_25_Lieto…

Il mio intervento di ieri su Radio Radicale, nella rubrica #PuntinisullAI di Marco Cerrone e incentrato "Intelligenza Artificiale e Astrazione: Metafore, ragionamento e pensiero invisibile" è disponibile qui: radioradicale.it/scheda/763661/… e su Spotify: open.spotify.com/episode/6SqAbM… #IA

Happy to share that our work "What Do Analogy Models Explain?", developed with Prof. Antonio Lieto @[email protected], has been accepted for presentation at #AISC2025, hosted by the IMT School for Advanced Studies, Lucca. Looking forward to exchanging ideas. See you in September! #CogSci #AI

La rivista Emerging Trends in Education ha pubblicato una intervista ad Antonio Lieto @[email protected], direttore del CIIT Lab, intitolata: La revolución tecnológica de la #inteligenciaartificial, sus límites y su potencial en la enseñanza revistaemerging.ujat.mx/index.php/emer… Università Salerno @PoliComUnisa

Hot take on OpenAI’s IMO gold What does it mean? I don’t know (yet). The fact that no tools, coding or internet was used is genuinely impressive. That said, my overall impression is that OpenAI has told us the result, but not how it was achieved. That leaves me with many

Terence Tao quietly seething at OpenAI is a sight to behold.

The smarter your AI is, the less data it needs. Needing huge training sets is a sign of stupidity.

PSA: Don't say "Longer thinking kills performance" when you mean "Length of intermediate token string is not correlated with final accuracy"

In my latest (and last!) column for Science’s Expert Voices series, I write about the reasons behind AI chatbots’ “deceptive” behaviors (and why Claude threatened a fictional CEO with blackmail). science.org/doi/10.1126/sc…

Who invented convolutional neural networks (CNNs)? 1969: Fukushima had CNN-relevant ReLUs [2]. 1979: Fukushima had the basic CNN architecture with convolution layers and downsampling layers [1]. Compute was 100 x more costly than in 1989, and a billion x more costly than

"A computer will do what you tell it to do, but that may be much different from what you had in mind." — Joseph Weizenbaum Image v/Joachim Schulz & Getty Images

LLMs are trained to imitate patterns of language, not to discover or verify truth. So, when asked to speak as an expert in an area where perceived experts have a widespread misconception, the LLM will parrot that misconception, adopting the register and vocabulary of experts.

Is Chain-of-Thought Reasoning of LLMs a Mirage? ... Our results reveal that CoT reasoning is a brittle mirage that vanishes when it is pushed beyond training distributions. This work offers a deeper understanding of why and when CoT reasoning fails, emphasizing the ongoing

My work here is truly done. Nobody with intellectual integrity can still believe that pure scaling will get us to AGI. GPT-5 may be a moderate quantitative improvement (and it may be cheaper) but it still fails in all the same qualitative ways as its predecessors, on chess, on

LLM reasoning is so wasteful it has erased neural networks’ efficiency advantage over symbolic AI.

The paper "#Emotion #Alignment in Human-Robot Interaction: Effects on Communication Styles and Persuasion" has been accepted in the IEEE Transactions on #AffectiveComputing (IF 9.8 ) and the early version is now available in IEEE Xplore doi.ieeecomputersociety.org/10.1109/TAFFC.… #HRI #socialrobots

I remember being excited about AI. I remember 20 years ago, being excited about neuroevolutionary methods for learning adaptive behaviors in video games. And I remember three years ago, mouth watering at the thought of tasty experiments in putting language models inside