DailyPapers

New survey on Continuous Diffusion and Flow Language Models is out! 🌊💬 One of my first papers, back in early 2025, was on continuous simplex flows. After working on discrete diffusion, I wanted to try controllable sequence generation with flow matching, which naturally operates in continuous spaces. But at the time, I found them much harder to train well than discrete diffusion. 🥲 2026 has changed this picture completely! 🚀 It’s been super exciting to watch continuous LMs take off, but with the rapid advancements, it’s also becoming harder to keep up with all the new work across pre-training, few-step generation, guidance, and scaling 🥹 So together with @chandavidlee, @FEijkelboom, and @jw9730, we put together a survey that consolidates the continuous LM literature and its key insights into 37 pages. It covers everything from the early work before the so-called “winter” of continuous LMs to the wave of new papers in 2026 🌟 More in thread! 🧵

yoco

Semantic Scholar (@SemanticScholar) is a rich source of information to find citation data of papers There's now an easy-to-use CLI and Skill so you can ask your agent things like "Find me all papers that cite the DeepSeek-R1 paper" github.com/huggingface/s2-cl…

Great to see so much interest in our post-training work. Want to reshare SAO (Single-Rollout Asynchronous Optimization), a key recent advance in agentic reinforcement learning alongside GLM-5.2. SAO tackles the off-policy and stability challenges of asynchronous RL through: - single-rollout sampling instead of GRPO-style group sampling - practical value-model training designs - a strict two-sided token-level clipping strategy SAO trains stably for 1,000 steps, consistently outperforms GRPO variants on agentic coding and reasoning benchmarks, and has been deployed in the agentic RL pipeline for GLM-5.2. arxiv.org/abs/2607.07508

Scaling self-verification with DeepSeek V4 Flash beats Claude Fable 5 on Terminal-Bench 2.1, while being 11x cheaper 💰 As open-source models become more capable, they can now generate large numbers of high-quality candidate solutions and verify their own outputs at very low cost. For example, we find that sampling just 5 solutions with DeepSeek V4 Flash and ranking them using the same model with LLM-as-a-Verifier can lead to a significant boost in accuracy (79% → 88%), outperforming closed frontier models on Terminal-Bench. Try it out today: github.com/llm-as-a-verifier… More on verification scaling in my previous post.