Daily AI news, sourcedUpdates / All
Updates 1960 items · newest issue first · page 39 of 98
AreaAllModelsAgents & codingApps & productsChips & infraResearchPricing & limitsBusiness & fundingPolicy & safetyOther
- Top 12 AI live translation tools compared on latency, languages and pricing, per MarkTechPostReportedMarkTechPostOct 12Apps & products
- Leading Chinese mathematicians unite behind use of AI in research, per South China Morning PostReportedSouth China Morning PostOct 12Research
- National Hurricane Center leaned on Google DeepMind's WeatherNext 3 for Hurricane Isaias, per CNN
- strix-llama rdna35-perf branch speeds Strix Halo prompt processing vs Vulkan and stock ROCm
- VectorDash is a Windows 11 local search app powered by EmbeddingGemma 2OfficialGitHub ampersandru/VectorDashOct 12Apps & products
- Bit-exact speculative decoding hits 106 t/s on Qwen3.8-27B with one RTX PRO 6000Unconfirmedr/LocalLLaMAOct 12Models
- Android receipt scanner runs a fine-tuned 0.8B vision model on-device for per-item grocery trackingUnconfirmedr/LocalLLaMAOct 12Apps & products
- SynAct helps LLM agents decide where scarce expert feedback will change the outcomeUnconfirmedarXiv cs.LGOct 12Research
- Paper treats multi-horizon event sequence forecasting as conditional generative modelingUnconfirmedarXiv cs.LGOct 12Research
- DiffusionART aims for efficient, stable RL fine-tuning of diffusion modelsUnconfirmedarXiv cs.LGOct 12Research
- Theory paper shows transformer masked diffusion models provably learn distributions with latent structureUnconfirmedarXiv cs.LGOct 12Research
- CacheScope speeds up diffusion transformers by choosing where to spend exact compute stepsUnconfirmedarXiv cs.LGOct 12Research
- GeopoliticalQA benchmarks RAG on 5,525 questions over 415 intelligence reportsUnconfirmedarXiv cs.CLOct 12Research
- Paper studies language models generating their own replay data to preserve capabilities during continual learningUnconfirmedarXiv cs.CLOct 12Research
- Study measures transferable gains versus prompt-level overfitting when optimizing a deep-research agent's promptsUnconfirmedarXiv cs.CLOct 12Research
- Elastic Latent Recurrence trains looped language models without the one-loop accuracy taxUnconfirmedarXiv cs.CLOct 12Research
- SWE-Duel proposes a self-scaling arena for coding agents that creates tasks at evaluation timeUnconfirmedarXiv cs.CLOct 12Research
- CLEAT trains interactive agents to ask clarifying questions guided by their decision valueUnconfirmedarXiv cs.AIOct 12Research
- Behavioral Self-Rewarding lets SWE agents judge their own debugging trajectories beyond pass/fail testsUnconfirmedarXiv cs.AIOct 12Research
- SpireBench tests 20 LLMs on full games of Slay the Spire for long-horizon strategyUnconfirmedarXiv cs.AIOct 12Research