Mengdi Wang

Looping Transformers can achieve better performance with shared parameter updates, revealing that scaling rules must adapt to parameter visits for stable recurrent depth.

Action-conditioned world model embeddings can revolutionize failure detection in long-horizon robotic tasks, achieving reliable monitoring without dense annotations.

EEVEE's innovative router-prompt co-evolution strategy enables LLMs to thrive in diverse, real-world task environments, outperforming existing methods by up to 48.2%.

Achieving 100% success on challenging theorem proving tasks, Goedel-Architect redefines efficiency in formal proofs by leveraging natural language guidance and adaptive blueprint refinement.

Success in long-horizon tasks hinges more on an agent's iterative persistence than on the quality of its initial solution.

Nemotron 3 Super proves you can achieve comparable accuracy to existing 120B models, but with significantly higher inference throughput, by combining Mamba, Attention, and Mixture-of-Experts.

Exact sampling in large-vocabulary decoding can be sped up by 19% simply by fusing it into the LM-head matmul, turning a bandwidth bottleneck into a lightweight epilogue.

AI agents that ace isolated coding tasks fall apart when faced with the messy reality of continuous software evolution, dropping from 80% to 38% success rates in a new benchmark.

Forget finetuning on curated datasets – OpenClaw-RL lets agents learn directly and continuously from *every* interaction, turning user replies, tool outputs, and even GUI changes into valuable RL signals.

LLMs still have a long way to go when it comes to interactive problem-solving, as revealed by a new benchmark that tests reasoning under budget constraints.