每天自动聚合 AI 领域最新动态
We study pretraining when compute is abundant but unique training data are finite, requiring repeated training on the...
Understanding and tracking a dynamic world are fundamental to visual intelligence. Existing video benchmarks provide ...
Omni-modal large language models (Omni-LLMs) extend LLMs to process heterogeneous modalities (e.g., vision, audio, an...
A robot following a multi-step instruction needs to act on the right objects in the right order, even as the scene ch...
Controlling an acting LLM agent means deciding, at each step of a long-horizon task, whether to trust its next action...
System One models choose an agent's next action by scoring a supplied set of options in a single pass. This makes the...
Off-policy data in reinforcement learning with verifiable rewards (RLVR) can arise from asynchronous generation, roll...
Reinforcement learning (RL) post-training improves robot foundation models (RFMs) beyond their demonstrations, but re...
Words are increasingly becoming an interface for action: large language models (LLMs) can now invoke tools, operate d...
Test-time scaling for large language models (LLMs) can involve generating multiple candidate solutions for selection ...
Latent diffusion models generate images as codes of an image tokenizer, whose decoder turns them into the pixel image...
Intent-based network management automates 6G service lifecycles, but existing fulfilment and assurance approaches req...