Kangwook Lee

Harness optimization can significantly enhance model performance, as demonstrated by WHALE's impressive accuracy gains across multiple domains.

LOTUS achieves a groundbreaking 2.5x-6.9x reduction in reasoning latency while matching explicit chain-of-thought performance at 3B parameters.

Asynchronous OPD can boost training throughput significantly while managing the challenges of stale data, transforming the efficiency of large language model fine-tuning.

RLDX-1 achieves double the success rate of existing VLAs on complex humanoid tasks, suggesting a leap in robots' ability to handle contact-rich, dynamic manipulation.

Even state-of-the-art language models struggle with exploration and exploitation in complex decision-making tasks, but minimal harness engineering can significantly improve performance.

LLM performance hinges on the code around the model, and Meta-Harness proves that automating the design of this "harness" can significantly boost results across diverse tasks.

Get faster long-context LLM inference without sacrificing accuracy: LookaheadKV predicts KV cache importance, outperforming costly draft generation methods by 14.5x.

Fine-tuning LLMs can kill their in-context learning abilities, but this work identifies a simple fix: only update the value matrix.

Language agents can now navigate complex, constrained environments with significantly improved success rates thanks to a new framework that combines multi-plan aggregation with constrained decoding and adaptive re-planning.