All Episodes

Position: The Pre/Post-Training Boundary Should Govern IP in Industry–Academia ML Collaborations

MEMO: Memory as a Model

Agent Bazaar: Enabling Economic Alignment in Multi-Agent Marketplaces

General Preference Reinforcement Learning

Explaining and Preventing Alignment Collapse in Iterative RLHF

Curriculum Learning-Guided Progressive Distillation in Large Language Models

Think Twice, Act Once: Verifier-Guided Action Selection For Embodied Agents

How Much Should a Conversational Recommender System Converse?

FUSE: Ensembling Verifiers with Zero Labeled Data

EVOLM: Self-Evolving Language Models through Co-Evolved Discriminative Rubrics

Personalized Alignment Revisited: The Necessity and Sufficiency of User Diversity

OGPO: Sample Efficient Full-Finetuning of Generative Control Policies

Adaptive Querying with AI Persona Priors

Rethinking the Role of LLMs in Time Series Forecasting

Robust Representation Learning through Explicit Environment Modeling

Magentic Marketplace: An Open-Source Environment for studying Agentic Markets

Hyperloop Transformers

Scaling Self-Play with Self-Guidance

RL Token: Bootstrapping Online RL with Vision-Language-Action Models

Agentic Data Environments

AI organizations are more effective but less aligned than individual agents

Text-to-Distribution Prediction with Quantile Tokens and Neighbor Context

Distortion of AI alignment revisited: RLHF is a decent utilitarian aligner

Llms get lost in multi-turn conversation

Transformers are inherently succint

The Coasean Singularity? Demand, Supply, and Market Design with AI Agents

Demystifying the unreasonable effectiveness of online alignment methods

Specialization after generalization: towards understanding test-time training in foundation models

Exploration and Exploitation Errors Are Measurable for Language Model Agents

A Mechanistic Analysis of Looped Reasoning Language Models

Sample Complexity of Autoregressive Reasoning: Chain-of-Thought vs. End-to-End

Why AI systems don’t learn and what to do about it

The Illusion of Learning from Observational Data: An Empirical Bayes Perspective

Ads in AI chatbots? An analysis of how large language models navigate conflicts of interest

Beyond Semantic Manipulation: Token-Space Attacks on Reward Models

LLM Evaluation as Tensor Completion: Low-Rank Efficiency and Uncertainty Quantification

Neural Computers

How AI Aggregation Affects Knowledge

World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry

In-Place Test-Time Training

Test-Time Scaling Makes Overtraining Compute-Optimal

AI Agent Prevalence and Data Quality Across Multiple Online Sample Providers

POLCA: Stochastic Generative Optimization with LLM

Agentic Markets: Equilibrium Effects of Improving Consumer Search

One Model, Two Markets: Bid-Aware Generative Recommendation

How Well Do LLMs Predict Human Behavior? A Measure of their Pretrained Knowledge

Learning to Reason with Curriculum I: Provable Benefits of Autocurriculum

Agentic AI and the next intelligence explosion

Understanding Behavior Cloning with Action Quantization

HyperAgents: : Open-Ended Metacognitive Self-Improvement for Any Computable Task

Harness design for long-running application development \ Anthropic

Reasonably reasoning AI agents can avoid game-theoretic failures in zero-shot, provably

How Log-Barrier Helps Exploration in Policy Optimization

The Finetuner’s Fallacy: When to Pretrain with Your Finetuning Data

TURNWISE: The Gap between Single- and Multi-turn Language Model Capabilities

Temporal Straightening for Latent Planning

Fine-Tuning Strategies for Preserving In-Context Learning in Linear Attention

LLMs Can Learn to Reason Via Off-Policy RL

Simple Recipe Works: Vision-Language-Action Models are Natural Continual Learners with Reinforcement Learning

Provable and practical in-context policy optimization for self-improvement

Matching Features, Not Tokens: Energy-Based Fine-Tuning of Language Models

Neural Thickets: Diverse Task Experts Are Dense Around Pretrained Weights

AdaEvolve: Adaptive LLM Driven Zeroth-Order Optimization

∇−reasoner: LLM reasoning via test-time gradient descent in latent space

Inference for Regression with Variables Generated by AI or Machine Learning

Fast KV Compaction via Attention Matching

Position: stop anthropomorphizing intermediate tokens as reasoning/thinking traces!

Code World Models for General Game Playing

Transformers Learn to Implement Multi-step Gradient Descent with Chain of Thought

Task Descriptors Help Transformers Learn Linear Models In-Context

Equivalence of Context and Parameter Updates in Modern Transformer Blocks

Learning without training: The implicit dynamics of in-context learning

Causal Identification from Counterfactual Data: Completeness and Bounding Results

Is Cosine-Similarity of Embeddings Really About Similarity?

Diffusion LLMs are Natural Adversaries for any LLM

Are you going to finish that? A Practical Study of the Partial Token Problem

Language Models Struggle to Use Representations Learned In-Context

LLMs are Bayesian, In Expectation, Not in Realization

Learning from Trials and Errors: Reflective Test-Time Planning for Embodied LLMs

LLMs Can Learn to Reason Via Off-Policy RL

Test-Time Training with KV Binding Is Secretly Linear Attention

Unified Latents (UL): How to train your latents

Spectral Bellman Method: Unifying RL Representation and Exploration

Prescriptive Scaling Reveals the Evolution of Language Model Capabilities

Experiential Reinforcement Learning

Learning Personalized Agents from Human Feedback

Learning to summarize user information for personalized RLHF

Intrinsic Credit Assignment for Long Horizon Interaction

Learning to Continually Learn via Meta-learning Agentic Memory Designs

Why Self-Rewarding Works: Theoretical Guarantees for Iterative Alignment of Language Models

PAD: Personalized Alignment of LLMs at Decoding-Time

The Reward Model Selection Crisis in Personalized Alignment

Causal-JEPA: Learning World Models through Object-Level Latent Interventions

How Sampling Shapes LLM Alignment: From One-Shot Optima to Iterative Dynamics

Deriving neural scaling laws from the statistics of natural language

Reasoning Cache: Continual Improvement Over Long Horizons via Short-Horizon RL

Scaling In-Context Online Learning Capability of LLMs via Cross-Episode Meta-RL

Divide-and-Conquer CoT: RL for Reducing Latency via Parallel Reasoning

Owning the AI Pareto Frontier — Jeff Dean

Learning to Reason in 13 Parameters