Can a 8B-parameter language model produce provably valid multi-step plans instead of plausible guesses? MIT CSAIL researchers introduce PDDL-INSTRUCT, an instruction-tuning framework that couples logical chain-of-thought with external plan validation (VAL) to lift symbolic planning performance of LLMs. On PlanBench, a tuned Llama-3-8B reaches 94% valid plans on Blocksworld, with large jumps on Mystery Blocksworld […]
The post MIT Researchers Enhanced Artificial Intelligence (AI) 64x Better at Planning, Achieving 94% Accuracy appeared first on MarkTechPost.