Daily AI news, sourcedUpdates / All
Updates 1960 items · newest issue first · page 51 of 98
AreaAllModelsAgents & codingApps & productsChips & infraResearchPricing & limitsBusiness & fundingPolicy & safetyOther
- Paper tests whether AI agents can do open-ended scientific discovery in the Station multi-agent environmentUnconfirmedr/MachineLearningOct 9Research
- Developer shows ALHR, a tree-based sparse attention method claiming 35x KV-cache read reductionUnconfirmedr/MachineLearningOct 9Research
- TRAE merges its Code and Work apps into one tool with Agent and IDE modes, per QbitAI
- Developer shares Memory Journal, a portable text-file format for carrying AI context and skills between chatsUnconfirmedr/LocalLLaMAOct 9Agents & coding
- LightOn releases LightOnOCR-3, a family of lightweight end-to-end OCR models led by a 4B model
- Recraft publishes a four-step guide to image-to-image generation in Recraft Studio
- Moonshot AI opens a Kimi Office beta that brings Kimi into WPS spreadsheets, slides and documents
- Developer releases MaRN, a PyTorch library that trains networks through compact low-dimensional parameter mappingsUnconfirmedr/MachineLearningOct 9Models
- Google deprecates the deep-research-pro-preview-12-2025 Deep Research agent in the Gemini API
- Tencent releases Youtu-Parsing-Omni, a 5B model that parses documents, charts, audio and video into structured JSON
- NVIDIA publishes a GR00T N1.7 deployment checkpoint for an Agile One S SSD-placement robot task
- A 100-object text-to-3D-game-object test ranks GPT-6 Astra most reliable while Opus 5.5's results look better, a Reddit post saysUnconfirmedr/singularityOct 9Models
- ByteDance Seed finds DeepSeek-V4's long-context recall swings with token position, per QbitAIReportedQbitAIOct 9Research
- Aether AI shows CRIS-0, a causal-reasoning robot system that replans within seconds of disruptions, per QbitAIReportedQbitAIOct 9Apps & products
- Lenovo's TianxiCode agent with DeepSeek-v4.1-Flash tops SWE-bench-Live Lite at 71%, per QbitAI
- DLoop paper proposes looped speculative decoding for faster LLM generationUnconfirmedr/LocalLLaMAOct 9Research
- Stepped MoE paper proposes segment-level routing with configurable inference costUnconfirmedr/LocalLLaMAOct 9Research
- Samsung expects a 780% jump in quarterly operating profit on AI-driven memory chip demand, per AFPReportedAFPOct 9Business & funding
- Free in-browser calculator estimates the electricity cost of running a local LLM server, developer saysUnconfirmedr/LocalLLaMAOct 9Apps & products
- Sophos cuts threat investigation time by 96% with OpenAI Daybreak agents