AI 速递 2026-07-21
生成时间：2026/7/21 09:24:26（UTC: 2026-07-21T01:24:26.091Z）
LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget
Section titled “LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget”👍 182 · arXiv
A growing gap separates inference context lengths from RL post-training: inference systems are approaching million-token contexts, while post-training workloads often remain at 256K tokens or below an…
VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding
Section titled “VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding”👍 155 · arXiv
Recent advances in video understanding have spanned motion, long video, and streaming interaction, driving this field toward real-world applications. Despite this progress, current open-source models …
RESOURCE2SKILL: Distilling Executable Agent Skills from Human-Created Multimodal Resources
Section titled “RESOURCE2SKILL: Distilling Executable Agent Skills from Human-Created Multimodal Resources”👍 114 · arXiv
Skills are a useful abstraction for software agents, turning human and agent experience into reusable procedural knowledge. Yet existing skill libraries are mostly hand-written, text-centric, or deriv…
RAGU: A Multi-Step GraphRAG Engine with a Compact Domain-Adapted LLM
Section titled “RAGU: A Multi-Step GraphRAG Engine with a Compact Domain-Adapted LLM”👍 111 · arXiv
Graph retrieval-augmented generation (GraphRAG) enhances large language models with structured knowledge, yet existing systems construct knowledge graphs in a single extraction pass, producing noisy e…
SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration
Section titled “SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration”👍 63 · arXiv
Recent advances in Tool-Integrated Large Language Models have made web search a core capability of information-seeking agents. However, as interaction histories grow, agents increasingly struggle to t…
Ollama v0.32.2-rc0
Section titled “Ollama v0.32.2-rc0”What’s Changed
Section titled “What’s Changed”- launch: keep Claude Code channels available by @hoyyeva in https://github.com/ollama/ollama/pull/17210
- cmd: remove dead agent prompt wrappers by @ParthSareen in https://github…
链接：https://github.com/ollama/ollama/releases/tag/v0.32.2-rc0
CrewAI 1.15.5
Section titled “CrewAI 1.15.5”What’s Changed
Section titled “What’s Changed”Features
Section titled “Features”- Authenticate skill registry downloads
Documentation
Section titled “Documentation”- Update snapshot and changelog for v1.15.4
Contributors
Section titled “Contributors”@vinibrsl…
链接：https://github.com/crewAIInc/crewAI/releases/tag/1.15.5
OpenAI Codex CLI rust-v0.145.0-alpha.25
Section titled “OpenAI Codex CLI rust-v0.145.0-alpha.25”Release 0.145.0-alpha.25
…
链接：https://github.com/openai/codex/releases/tag/rust-v0.145.0-alpha.25
Agent swarms and the new model economics
Section titled “Agent swarms and the new model economics”Article URL: https://cursor.com/blog/agent-swarm-model-economics Comments URL: https://news.ycombinator.com/item?id=48982535 Points: 117
Comments: 48
Section titled “Comments: 48”How we measured AI writing across arXiv, and where the measurement breaks
Section titled “How we measured AI writing across arXiv, and where the measurement breaks”Article URL: https://unslop.run/blog/measuring-ai-writing-on-arxiv Comments URL: https://news.ycombinator.com/item?id=48981206 Points: 192
Comments: 140
Section titled “Comments: 140”Launch HN: Bloomy (YC S26) – AI-powered mastery learning for K-12
Section titled “Launch HN: Bloomy (YC S26) – AI-powered mastery learning for K-12”Hi HN, I’m Alex Southmayd, the founder of Bloomy (https://bloomylearning.com) – an AI-powered mastery-learning platform for K-12 students. Bloomy provides students with an AI tutor alongside adaptive curriculum (right now Math, English Language Arts, and Writing).How it works: we diagnose students’
Mythologizing AI makes it more likely that we’ll fail to operate it well (2023)
Section titled “Mythologizing AI makes it more likely that we’ll fail to operate it well (2023)”Article URL: https://www.newyorker.com/science/annals-of-artificial-intelligence/there-is-no-ai Comments URL: https://news.ycombinator.com/item?id=48980238 Points: 60
Comments: 97
Section titled “Comments: 97”China’s open-weights AI strategy is winning
Section titled “China’s open-weights AI strategy is winning”Article URL: https://werd.io/american-ai-is-locked-down-and-proprietary-its-losing/ Comments URL: https://news.ycombinator.com/item?id=48979269 Points: 974
Comments: 776
Section titled “Comments: 776”Claude Fable produced a counterexample to the Jacobian Conjecture
Section titled “Claude Fable produced a counterexample to the Jacobian Conjecture”Article URL: https://xcancel.com/__alpoge__/status/2079028340955197566 Comments URL: https://news.ycombinator.com/item?id=48973869 Points: 666
Comments: 428
Section titled “Comments: 428”Ask HN: What are your favorite blogs not about AI?
Section titled “Ask HN: What are your favorite blogs not about AI?”Many of my favorite bloggers have stopped writing (or have drastically shifted topics). What are your favorite blogs?I miss the 2009 era of blogging about personal projects or super niche factoids.
Comments URL: https://news.ycombinator.com/item?id=48972858 Points: 87
Comments: 37
Section titled “Comments: 37”AI advice made people less accurate but more confident – sudy
Section titled “AI advice made people less accurate but more confident – sudy”Article URL: https://thenextweb.com/news/ai-advice-suppresses-critical-thinking-wrong-answers-study Comments URL: https://news.ycombinator.com/item?id=48971738 Points: 357