人人能懂的AI前沿解读(10.18)

[LG] Are We Overthinking One-Step Generation?
[Seoul National University & Harvard University]
https://arxiv.org/abs/2610.20331
---
[LG] LLM-to-Flow Distillation: Teaching Natural Language Goals to Scientific Generators
[ETH Zurich & Technische Universität Wien]
https://arxiv.org/abs/2610.17246
---
[AI] Scalable Weak-to-Strong Verifier Hardening for Agent Environments
[Stanford University & CMU & Fewshot Corp]
https://arxiv.org/abs/2610.15924
---
[CL] DuoDiscover: Scaling Autonomous Discovery Beyond the Evaluation Bottleneck
[Google]
https://arxiv.org/abs/2610.17184
---
[LG] Training Language Models to Fail Gracefully
[CMU & UC Berkeley]
https://arxiv.org/abs/2610.20401