Xiaoran Liu

Hybrid LLMs exhibit a "Seesaw Effect" in context extension, with linear attention models thriving on continual pretraining and sliding-window models excelling at extrapolation, a distinction rooted in their positional inductive biases.

Automating the creation of RL environment families could drastically reduce development time and enhance agent training efficiency.

Forget benchmarks: AI can now learn "scientific taste" and propose research ideas with higher potential impact than humans, thanks to a novel reinforcement learning approach using citation data.

Simulation offers a systematic path to overcoming data bottlenecks in AI, but designing effective digital twins requires careful consideration.

Digital twins promise safer, cheaper, and more automated smart grids by providing high-fidelity virtual replicas for design, testing, and real-time operation.