Cristiano Codagnone’s Post

Many thanks to the team from DEF for triggering this insightful report about critical hinking and generative AI. I am afraid that it will be of increasing importance in the future.

Cristiano Codagnone 's report is widely researched and presents four possible scenarios that are worth your consideration. #GenAI #criticalthinking #labour #juniorjobs

Excited to share that our work has been accepted to ICML 2026! 🎉 DEGAP: Dynamic Entropy-Guided Attention Perturbation for Contrastive Decoding in Large Vision-Language Models Our work, DEGAP, introduces a training-free method for reducing hallucinations in Large Vision-Language Models by dynamically selecting the optimal attention masking layer through entropy-guided analysis. Looking forward to presenting our work at ICML and meeting researchers working on multimodal AI! Thanks to my collaborators and advisor for making this work possible. #ICML2026 #AI #VLM #LLM #ComputerVision #MultimodalAI

We are excited for the workshop programme at #ICML 2026 to begin tomorrow. Six papers involving members of our lab are included in tomorrow’s workshop programme, across generative modelling, technical AI governance, probabilistic inference, agentic AI and decision-making. The workshop programme offers a valuable opportunity to explore emerging ideas and open questions across these rapidly developing areas. We are looking forward to seeing these ideas contribute to discussions across the ICML workshop programme.

$0 to $100M ARR in 8 Months. I sat down with Anastasios Angelopoulos to understand how Arena became the neutral judge of the AI race. It started as a research project at UC Berkeley, with PhD students asking a simple question: are we measuring AI the right way? Today, Arena is the largest real-world AI evaluation platform in the world, reaching 10M monthly users in 150 countries and 700M conversations We get into: - Why traditional AI benchmarks are fundamentally broken - How Arena's leaderboard became the industry standard - Agent Mode: millions of agentic traces, causal tracing - How humans are actually using AI agents today - From Berkeley research project to $100M ARR FULL CONVERSATION LINKED BELOW

On episode 2 of Lab Notes, Amir Zohrenejad and Hanchen Li explore the systems that make modern AI agents faster, more efficient, and better at learning from experience. They discuss KV Cache optimization, long-context inference, prompt learning, continual learning, and why better benchmarks may be just as important as larger models in advancing AI research. Tune in! https://hubs.ly/Q04nKZfd0

At The Japanese Society for Artificial Intelligence 2026, a session on Collective Predictive Coding and the Era of New Intelligence brought together Tadahiro Taniguchi, Yusuke Hayashi, Ken Suzuki to explore how Collective Predictive Coding can help us rethink intelligence in the age of AI. The discussion explored how Collective Predictive Coding can help us rethink intelligence in the age of AI — not as something possessed by individual agents alone, but as something that emerges through coordination, communication, and shared meaning-making. The session connected CPC to three urgent challenges: democracy, science, and temporal scales. Read the full event report here: https://lnkd.in/dN4qYHaR 

Amy Allhouse and Emily Haikney from University of Leeds are at #ALDcon26 facilitating their workshop on Navigating Skill Development in the Era of Generative AI – international students' perspectives #LoveLD https://aldinhe.ac.uk/

Harnessing human capital is essential for amplifying our expertise. By building learning loops on top of existing models, we aim to outpace innovations and elevate our journey with AI. This approach will enable us to continuously adapt and thrive in an ever-evolving landscape.

Is AI conscious or just a calculator? If you missed this thought provoking webinar and feel both fascinated and conflicted about AI, it is well worth your time. You can watch the recording here: https://accntu.re/3QnDOTE This is not a webinar where we demonstrate tools or discuss responsible AI frameworks. Using intuitive ideas from quantum mechanics and lessons from history, Dr. Dan Grahn looks at why some systems resist clean explanations, why confidence can be misleading, and what it means to approach AI with more humility and care.