NeurIPS paper reviews 2025 #3

Discover the perspectives of Nick, one of our Quantitative Researchers, on the following papers:

• Counterfactual Identifiability via Dynamic Optimal Transport
• Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?
• Progressive Data Dropout: An Embarrassingly Simple Approach to Train Faster

Discover the perspectives of Tomi, one of our Quantitative Research Managers, on the following papers:

• Kronos: A Foundation Model for the Language of Financial Markets
• LOBERT: Generative AI Foundation Model for Limit Order Book Messages
• Auto-Compressing Networks

Discover the perspectives of Nick, one of our Software Engineers, on the following papers:

• Pass@K Policy Optimisation: Solving Harder Reinforcement Learning Problems
• Antidistillation Sampling
• Fundamental Limitations in Pointwise Defences of LLM Finetuning APIs

Discover the perspectives of Cédric, one of our Quantitative Researchers, on the following papers:

• Learning (Approximately) Equivariant Networks via Constrained Optimisation
• On the Closed-Form of Flow Matching: Generalization Does Not Arise from Target Stochasticity

Discover the perspectives of Ognjen, one of our Quantitative Researchers, on the following papers:

• Omnipresent Yet Overlooked: Heat Kernels in Combinatorial Bayesian Optimisation
• Bubbleformer: Forecasting Boiling with Transformers

Discover the perspectives of Radomir, one of our Machine Learning Engineers, on the following papers:

• Learning Task-Agnostic Representations through Multi-Teacher Distillation
• Contrastive Representations for Temporal Reasoning

Discover the perspectives of Benjamin, one of our Quantitative Researchers, on the following papers:

• Dynamical Decoupling of Generalisation and Overfitting in Large Two-Layer Networks
• Backward Conformal Prediction
• Predicting the Performance of Black-box Language Models with Follow-up Queries

Discover the perspectives of Casey, one of our Machine Learning Engineers, on the following papers:

• Distributed Orthonormal Updates for Large-Scale Training
• Broken Tokens? Your Language Model can Secretly Handle Non-Canonical Tokenisations

Discover the perspectives of Hugh, one of our Quantitative Research Managers, on the following papers:

• Understanding and Mitigating Numerical Sources of Non-determinism in LLM Inference
• Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free

Discover the perspectives of Timothy, one of our Machine Learning Engineers, on the following papers:

• ZeroS: Zero-Sum Linear Attention for Efficient Transformers
• In Search of Adam’s Secret Sauce

Discover the perspectives of David, one of our Quantitative Researchers, on the following papers:

• 1000 Layer Networks for Self-Supervised RL: Scaling Depth Can Enable New Goal-Reaching Capabilities
• Revisiting Residual Connections: Orthogonal Updates for Stable and Efficient Deep Networks

Discover the perspectives of Szymon, one of our Senior Quantitative Researchers, on the following papers:

• Rollout Roulette: A Probabilistic Inference Approach to Inference-Time Scaling of LLMs using Particle-Based Monte Carlo Methods
• Parallelizing MCMC Across the Sequence Length

Discover the perspectives of Simon, one of our Senior Quantitative Researchers, on the following papers:

• Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond)
• ConfTuner: Training Large Language Models to Express Their Confidence Verbally