Latest

Every published item, in order · 9845 items
Oct 27
Tomorrow 08:00, the daily lands in your inboxTelegram channel or email. Same screening as this page, no ads.
PapersBuildersSelected10:24
Speculative Decoding for MoE Models: Deduplication and Draft Length

Deduplicating expert reads sounds appealing for MoE inference, but the measured bandwidth numbers show it rarely helps at practical draft lengths, which saves practitioners from engineering it blindly. Inference engineers and MoE serving teams can reuse the fitted-bandwidth method, which takes under an hour per model, to predict whether deduplication wins on their own hardware. The tested stacks — llama.cpp CUDA and Metal plus MLX-LM — cover the common Mac and NVIDIA deployment paths.

arXiv: DeepSeek@Yuzheng LanSource
PapersBuildersSelected09:38
AutoTavern: LLM Agents Build Hearthstone Battlegrounds AI Rivaling Top Players

The headline result is less the game AI itself than the demonstration that LLM agents constructing and verifying research infrastructure outperform those same LLMs playing directly. Practitioners building agent pipelines can study its typed research memory, verification gates and feedback loops as a template for long-horizon autonomous work. The open-sourced TavernGym and TavernBench also give anyone a ready-made RL environment with stochastic, hidden-information decision-making.

arXiv cs.AI@Changlian Ma, Limin WangSource

Only the latest 2000 items are listed; for older ones see the daily archive