The Berkeley Artificial Intelligence Research Blog

4 articles in the TruthFoundry index. Each links out to the original.

• ABBEL Framework Uses Belief States to Improve Long-Horizon LLM Coding · ABBEL framework uses supervised belief states to reduce context costs in LLM coding assistants.
• Berkeley Researchers Propose Redesigning Data Systems for the Era of Free AI Agents · UC Berkeley researchers outline three challenges for data systems as AI inference costs approach zero.
• Berkeley AI Lab Celebrates 2026 Ph.D. Graduates in Robotics and LLMs · BAIR Lab honors 2026 Ph.D. graduates advancing AI in robotics, language models, and safety.
• Adaptive Parallel Reasoning: New Paradigm for Efficient LLM Inference Scaling · Researchers introduce adaptive parallel reasoning to let models dynamically control inference scaling.