Dr François Pachet is an AI researcher and musician, and one of the most influential figures in AI and music. His innovative contributions have defined the field over the past decades through creative systems such as the Continuator, and Flow Machines, among others. After leading the Spotify Creator Technology Research Lab and the Sony Computer Science Lab, he went on to create his own companies: Imagine All The People and Ynosound. In the context of IJCAI2025, he spoke about what deep learning changed, and what still remains wide open. He explains why tools like Suno and Udio--ChatGPT-like platforms for music generation--can produce astonishing results that still feel unsatisfying; why the next step for music generation requires combining sampling with search; and why the most important problems in artistic domains are, by nature, ill-defined--because there is no loss function to determine what is "good". Above all, he defends the importance of researcher autonomy: work on the questions that genuinely fascinate you, even when they fall outside prevailing trends--perhaps especially then. Thank you for joining me for this interview. Could you begin by telling us when was your first IJCAI and a memory related to it? I think the first IJCAI I attended was in Montreal in '95. I was there for a couple of workshops, one of them was about music and AI, and the other one I think was on advisor systems, something like that. And I remember there was a French colleague who was there also, at the time he was doing his PhD. And there was this researcher called Herbert Simon, who is a Nobel Prize pioneer of AI. I remember having chatted a little bit with this French guy who was very bold, and he just went up to Simon, said "Hey," and he started a conversation with him. And I was very impressed by the fact that you could meet those kinds of guys informally in a corridor or something at this conference.

Can Michael Pollan crack the problem of consciousness in his new book? It is one of the most perplexing questions in science. You would expect our intimacy with it to give us a leg up in understanding how it works, but this has proven to be more of a hindrance than a help. So how can you study something objectively when it is also the very tool you are using to do the studying? This conundrum forms the backbone of Michael Pollan's latest book, Pollan's previous works include and The former helped bring the environmental and animal welfare impacts of the US food system to light, while the latter introduced the public to the psychedelic research renaissance.

Neural Information Processing Systems http://nips.cc/

Deep neural networks are powerful machines for visual pattern recognition, but reasoning tasks that are easy for humans may still be difficult for neural models. Humans possess the ability to extrapolate reasoning strategies learned on simple problems to solve harder examples, often by thinking for longer. For example, a person who has learned to solve small mazes can easily extend the very same search techniques to solve much larger mazes by spending more time. In computers, this behavior is often achieved through the use of algorithms, which scale to arbitrarily hard problem instances at the cost of more computation. In contrast, the sequential computing budget of feed-forward neural networks is limited by their depth, and networks trained on simple problems have no way of extending their reasoning to accommodate harder problems. In this work, we show that recurrent networks trained to solve simple problems with few recurrent steps can indeed solve much more complex problems simply by performing additional recurrences during inference. We demonstrate this algorithmic behavior of recurrent networks on prefix sum computation, mazes, and chess. In all three domains, networks trained on simple problem instances are able to extend their reasoning abilities at test time simply by thinking for longer.

Large Language Models (LLMs) have demonstrated impressive capabilities in natural language and code generation, and are increasingly deployed as automatic judges of model outputs and learning activities. Yet, their behavior on structured tasks such as predicting the difficulty of competitive programming problems remains under-explored. We conduct a systematic comparison of GPT-4o, used purely as a natural-language difficulty assessor, against an interpretable Light-GBM ensemble trained on explicit numeric and textual features. On a dataset of 1,825 LeetCode problems labeled Easy, Medium, or Hard, LightGBM attains 86% accuracy, whereas GPT-4o reaches only 37.75%. Detailed analyses, including confusion matrices and SHAP-based interpretability, show that numeric constraints -- such as input size limits and acceptance rates -- play a crucial role in separating Hard problems from easier ones. By contrast, GPT-4o often overlooks these cues and exhibits a strong bias toward simpler categories. We further probe GPT-4o through a synthetic Hard-problem generation protocol. Surprisingly, GPT-4o labels almost all of its own synthetic Hard problems as Medium, contradicting its tendency to downgrade real Hard problems to Easy. Our findings connect to recent work on LLMs-as-judges and automatic difficulty estimation in programming and education, and highlight concrete failure modes that must be addressed before LLM-based judges can be considered trustworthy in competitive programming, educational platforms, or reinforcement-learning pipelines.

Neural Information Processing Systems http://nips.cc/

This blog post is based on the work BaNEL: Exploration Posteriors for Generative Modeling Using Only Negative Rewards . The ultimate aim of machine learning research is to push machines beyond human limits in critical applications, including the next generation of theorem proving, algorithmic problem solving, and drug discovery. A standard recipe involves: (1) pre-training models on existing data to obtain base models, and then (2) post-training them using scalar reward signals that measure the quality or correctness of the generated samples. The probability of producing a positive-reward sample can be so low that the model may go through most of the training without ever encountering a positive reward. Calls to the reward oracle can be expensive or risky, requiring costly simulations, computations, or even physical experiments.

One Hand Clapping covers a lot of ground, which can make it seem like an entertaining lecture series, with amusing sketches. Some may find Kukushkin's playfulness a bit much.

Long-form chain-of-thought reasoning has become a cornerstone of advanced reasoning in large language models. While recent verification-refinement frameworks have enabled proprietary models to solve Olympiad-level problems, their effectiveness hinges on strong, reliable verification and correction capabilities, which remain fragile in open-weight, smaller-scale models. This work demonstrates that even with weak verification and refinement capabilities on hard tasks, the reasoning limits of such models can be substantially extended through a probabilistic paradigm we call Deep Self-Evolving Reasoning (DSER). We conceptualize iterative reasoning as a Markov chain, where each step represents a stochastic transition in the solution space. The key insight is that convergence to a correct solution is guaranteed as long as the probability of improvement marginally exceeds that of degradation. By running multiple long-horizon, self-evolving processes in parallel, DSER amplifies these small positive tendencies, enabling the model to asymptotically approach correct answers. Empirically, we apply DSER to the DeepSeek-R1-0528-Qwen3-8B model. On the challenging AIME 2024-2025 benchmark, DSER solves 5 out of 9 previously unsolvable problems and boosts overall performance, enabling this compact model to surpass the single-turn accuracy of its 600B-parameter teacher through majority voting. Beyond its immediate utility for test-time scaling, the DSER framework serves to diagnose the fundamental limitations of current open-weight reasoners. By clearly delineating their shortcomings in self-verification, refinement, and stability, our findings establish a clear research agenda for developing next-generation models with powerful, intrinsic self-evolving capabilities.

Reinforcement learning (RL) has emerged as a central paradigm for training large language models (LLMs) in reasoning tasks. Yet recent studies question RL's ability to incentivize reasoning capacity beyond the base model. This raises a key challenge: how can RL be adapted to solve harder reasoning problems more effectively? To address this challenge, we propose a simple yet effective strategy via Question Augmentation: introduce partial solutions during training to reduce problem difficulty and provide more informative learning signals. Our method, QuestA, when applied during RL training on math reasoning tasks, not only improves pass@1 but also pass@k-particularly on problems where standard RL struggles to make progress. This enables continual improvement over strong open-source models such as DeepScaleR and OpenMath Nemotron, further enhancing their reasoning capabilities. We achieve new state-of-the-art results on math benchmarks using 1.5B-parameter models: 72.50% (+10.73%) on AIME24, 62.29% (+12.79%) on AIME25, and 41.67% (+10.11%) on HMMT25. Code, data and model are available at https://github.com/foreverlasting1202/QuestA.