AI 速递 2026-07-26
生成时间：2026/7/26 09:30:13（UTC: 2026-07-26T01:30:13.831Z）
AREX: Towards a Recursively Self-Improving Agent for Deep Research
Section titled “AREX: Towards a Recursively Self-Improving Agent for Deep Research”👍 127 · arXiv
Deep research requires agents to find answers that jointly satisfy multiple constraints. Discovering such answers is costly, whereas verifying a candidate can often be decomposed into tractable constr…
SLAI T-Rex: Full-Parameter Post-training of the DeepSeek-V4 Family on Ascend SuperPOD
Section titled “SLAI T-Rex: Full-Parameter Post-training of the DeepSeek-V4 Family on Ascend SuperPOD”👍 56 · arXiv
Full-parameter post-training of trillion-parameter-scale MoE models introduces substantial system-level challenges for large-scale distributed training, including severe memory pressure, non-overlappe…
K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs
Section titled “K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs”👍 55 · arXiv
Large language models are increasingly used in K-12 education, but existing benchmarks mainly test exam question answering rather than understanding how curriculum knowledge is structured and visually…
ReferTrack: Referring Then Tracking for Embodied Visual Tracking
Section titled “ReferTrack: Referring Then Tracking for Embodied Visual Tracking”👍 49 · arXiv
Embodied visual tracking (EVT) requires a mobile agent to continuously follow a specific target described in natural language using only onboard vision. While recent vision-language-action (VLA) polic…
Visual Contrastive Self-Distillation
Section titled “Visual Contrastive Self-Distillation”👍 41 · arXiv
On-policy self-distillation (OPSD) is promising as it removes the external teacher required by on-policy distillation (OPD), yet it still needs asymmetric information between teacher and student to en…
LangChain langchain-anthropic==1.5.2
Section titled “LangChain langchain-anthropic==1.5.2”Changes since langchain-anthropic==1.5.1
release(anthropic): 1.5.2 (#39056) feat(anthropic): add Claude Opus 5 support (#39054)…
链接：https://github.com/langchain-ai/langchain/releases/tag/langchain-anthropic%3D%3D1.5.2
vLLM v0.26.0
Section titled “vLLM v0.26.0”vLLM v0.26.0 Release Notes
Section titled “vLLM v0.26.0 Release Notes”Highlights
Section titled “Highlights”This release features 411 commits from 212 contributors (61 new)!
- New Inkling model family with a full support stack: base modeling (#48799), …
链接：https://github.com/vllm-project/vllm/releases/tag/v0.26.0
Ollama v0.32.4
Section titled “Ollama v0.32.4”What’s Changed
Section titled “What’s Changed”- Support Laguna on Apple GPUs via the MLX engine
- Quantize draft-model output heads at the requested type when creating speculative-decoding drafts.
- Fixed Qwen3 MoE decoding f…
链接：https://github.com/ollama/ollama/releases/tag/v0.32.4
CrewAI 1.15.6
Section titled “CrewAI 1.15.6”What’s Changed
Section titled “What’s Changed”Bug Fixes
Section titled “Bug Fixes”- Fix detection of Anthropic preview tool-use blocks.
- Preserve strict tool schema property names.
- Dispatch execution_end hook on failed crew and flow executions. -…
链接：https://github.com/crewAIInc/crewAI/releases/tag/1.15.6
OpenAI Codex CLI rust-v0.146.0-alpha.10.1
Section titled “OpenAI Codex CLI rust-v0.146.0-alpha.10.1”Release 0.146.0-alpha.10.1
…
链接：https://github.com/openai/codex/releases/tag/rust-v0.146.0-alpha.10.1
Librarians are hosting viral ‘Avoiding AI’ workshops for people who are fed up with Big Tech
Section titled “Librarians are hosting viral ‘Avoiding AI’ workshops for people who are fed up with Big Tech”At libraries around the country, “Avoiding AI” workshops have elicited unprecedented demand.
One fallen power line exposed a growing AI data center problem. Here’s how to fix it.
Section titled “One fallen power line exposed a growing AI data center problem. Here’s how to fix it.”A close call in Northern Virginia revealed just how poorly data centers respond to grid disruptions. Here’s how to fix the problem.
I tried out OpenAI’s new AI keypad — which will be fun for some coders and slightly mystifying to everyone else
Section titled “I tried out OpenAI’s new AI keypad — which will be fun for some coders and slightly mystifying to everyone else”OpenAI’s fancy new AI keypad will be a lot of fun for some, while many others are probably not going to touch it.
Prentis, new AI lab co-founded by Reid Hoffman, Mark Pincus in talks to raise $100M
Section titled “Prentis, new AI lab co-founded by Reid Hoffman, Mark Pincus in talks to raise $100M”The neolab is betting that automating routine computer tasks will soon outpace coding as AI’s biggest use case.
Why Cognition bought Poke: AI personality is becoming a competitive advantage
Section titled “Why Cognition bought Poke: AI personality is becoming a competitive advantage”The acquisition brings Poke’s conversational style and interaction model to Cognition’s coding agent Devin, reflecting a growing belief that how AI assistants interact with users is as important as the models powering them.
Anthropic launches Opus 5
Section titled “Anthropic launches Opus 5”Opus 5 will be both cheaper and less restrictive than Fable, likely making it preferable in most use cases.
As US weighs response to Chinese AI, industry urges against broad open-weight restrictions
Section titled “As US weighs response to Chinese AI, industry urges against broad open-weight restrictions”AI companies, including Nvidia and Mistral, urge policymakers to avoid broad restrictions on open-weight AI models as Washington debates responses to Chinese AI and alleged model distillation.
Bluesky’s AI assistant Attie expands into an open social research tool
Section titled “Bluesky’s AI assistant Attie expands into an open social research tool”Users can now ask Attie questions about news, trends, and conversations on Bluesky and other apps on the AT Protocol.