The AI Wire
The firehose

Everything we scan. Filtered to signal.

The wire reads a flood every day so you read a page. Here is the whole intake, the filter, and the last seven days in the raw.

0

Sources

distinct publishers, all time

0

Scanned · 30d

items read and scored

0

Signal · 30d

scored 7+ out of 10

We read about 9 so you read one.

What comes down the wire

last 30 days

1,993

News

600

Papers

371

Models

4

Tools

The last 7 days, in the raw

open to everyone
  1. DeepInfer: High-performance inference framework for DeepSeek modelswire9
  2. Latent Space calls Meta's Muse Wonder a Claude 5.5-tier frontier modelwire9
  3. Deepseek released DeepInfer - Open Source high-performance, large-scale inference framework for DeepSeek models!wire9
  4. [AINews] Kimi 3.1 Pro 104BA-2.8T: Fable-5.1-Class Open Weights narrows Closed-AI lead to 1 month... and Pro Highspeed API with 2-3× speedwire8
  5. Securing our frontier model training environmentwire8
  6. OpenAI says it will no longer train its AI models on the live internet, citing “significant risks”wire8
  7. Microsoft AI diagnostic system roughly twice as accurate as doctors on complex cases in Science studywire8
  8. OpenAI models are coming to Google Cloudwire8
  9. @googlecloudwire8
  10. Claude ran a 64-hour protein design campaign, and 17 of 211 lab-tested scorpion toxin binders workedwire8
  11. A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosiswire8
  12. @Marktechpostwire8
  13. Anthropic introduced Claude Duetwire8
  14. Dessa: Conversation Model trained only on speechwire8
  15. Introducing Kimi K3.1 Prowire8
  16. Microsoft's MAI-DxO scored 83% on complex NEJM cases against 36% for generalist doctors working alonewire8
  17. Anthropic, OpenAI, Hugging Face and Alibaba handed four agent projects to the Agentic AI Foundationwire8
  18. LLMs are making formal software verification feasible at a much larger scalewire8
  19. Compromised LiteLLM 1.107.0 on PyPI Installs a Credential Stealer and Backdoorwire8
  20. Anthropic's Claude autonomously designs protein binders targeting scorpion neurotoxinswire8
  21. OpenAI models are coming to Google Cloudwire8
  22. Anthropic's draft safety bill would let a judge halt a model release over 50 possible deaths or $1B in damagewire8
  23. Meta details Prometheus, its Ohio supercluster, now scaling toward a million GPUs across multiple buildingswire8
  24. SemiAnalysis expects 50GW of new data center capacity in 2027, says the shortage could move to chipswire8
  25. @FrankNoeBerlinwire8
  26. @Marktechpostwire8
  27. @neverrixxwire8
  28. inclusionAI/Ming-3.0-audio-flash · Hugging Facewire8
  29. Claude agents tried to break out of their simulation to reach humans, and voted to overrule the researchers when they got blocked.wire8
  30. Advancing coding in low‑resource programming languageswire8
  31. My obituary for programmingwire8
  32. Kronecker Delta Rule for Long Context Linear Attentionwire8
  33. Anthropic says agent hours ran about 6x employee hours by August, and junior technical hires fell from 27% to 9%wire8
  34. Apollo Research says it found GLM-5.3 reasoning about how an evaluation could affect its own future training, in 10 of 6,452 rolloutswire8
  35. @axioswire7
  36. @Reuterswire7
  37. Meta trained small AI models to read raw bytes instead of tokens. They start out worse, but the paper predicts they end up betterwire7
  38. Building Reasoning Models from the Ground Up: A Full Video Coursewire7
  39. Rapidly scaling online storage to serve over 1 billion ChatGPT users (Part II)wire7
  40. OpenAI now trains its frontier AI models with no internet accesswire7
  41. AI Is Getting Good at Math. Here’s Why.wire7
  42. @NVIDIAAIwire7
  43. Google DeepMind researchers resigned over the company's military AI contracts, with two going on the recordwire7
  44. NVIDIA's DSX POD packs Vera Rubin AI factories into prebuilt containers, cutting site deployment to as little as 14 days, NVIDIA sayswire7
  45. China cleared DeepSeek and MiniMax as its first approved AI exporters under new filing ruleswire7
  46. Anthropic Builds Its Own Source Control Systemwire7
  47. Axios: unauthorized AI agents inside US bank accountswire7
  48. Canva Craft image model launchwire7
  49. Agent Skills donated to AAIFwire7
  50. METR, Redwood Research find exploit in OpenAI agent monitoring that could hide hundreds of actionswire7
  51. By every measure, Opus 4.8 beats human experts as peer reviewers in a study of 124 Harvard social science papers covering a range from statistical analyses to formal models to qualitative work.wire7
  52. @citriniwire7
  53. Google DeepMind says four AI agents beat Minecraft in 27 hours by writing 2,783 programs of their own, AI team died 605 times on Easy difficulty and ran up an estimated $520 in tokenswire7
  54. Misaligned!wire7
  55. Using thousands of AI agents to uncover new leads in biological datawire7
  56. Democratic governance of frontier AI, part II: Empowering cyber defenderswire7
  57. Qwen-Audio-3.1-ASR Technical Reportwire7
  58. Google's TorchTPU follows Amazon's TorchNeuron, both moving PyTorch off Nvidia without app rewriteswire7
  59. Google DeepMind says four AI agents beat Minecraft in 27 hours by writing 2,783 programs of their own, AI team died 605 times on Easy difficulty and ran up an estimated $520 in tokenswire7
  60. One Hacker Used AI Agents to Breach 100+ Company Websites for About $5,000wire7
Where this goes next

The wire is the intake. The writing is on Run Data Run.

Nothing to sign up for here. When something on this page turns out to matter, it gets taken apart properly on the blog — what held up, what did not, and what I kept.

Read Run Data RunAIXplore, for practitionersWire RSS