HN Radio.daily Hacker News, read aloud

← all episodes

Machine Learning & AI

70 stories across the archive · newest first

most mentionedAnthropicOpenAIClaudeClaude CodeGLM-5.2CodexAIChatGPTClaude Fable 5Cursor
  1. OpenAI ships GPT-5.6 with a cheaper, multi-model push▲ 1529 · 1087 commentsin 2 episodes:▶ · 0:53▶ · 0:35
  2. Grok 4.5 arrives as xAI’s coding-focused model, with Cursor in the loop▲ 771 · 1472 comments▶ · 2:08
  3. Tencent ships Hy3 as a cheaper open model contender▲ 553 · 117 comments▶ · 2:32
  4. OpenAI turns ChatGPT into a work agent, and HN hates the app shuffle▲ 348 · 186 comments▶ · 2:57
  5. Meta ships Muse Spark 1.1 and opens a new model API▲ 405 · 211 comments▶ · 3:23
  6. Colibrì gets a 744B-parameter model crawling on consumer hardware▲ 871 · 215 comments▶ · 3:49
  7. A beginner-friendly ML paper list goes viral, with caveats▲ 640 · 110 commentsin 2 episodes:▶ · 0:37▶ · 2:33
  8. Kokoro makes local text-to-speech feel practical on ordinary CPUs▲ 513 · 97 commentsin 2 episodes:▶ · 1:01▶ · 3:16
  9. OpenAI sets Thursday launch for GPT-5.6 Sol, Terra, and Luna▲ 235 · 210 comments▶ · 1:28
  10. Claude extends Fable 5 access, but users complain about quota whiplash▲ 231 · 250 comments▶ · 1:56
  11. Open-weight AI models threaten the inference-margin story▲ 690 · 468 comments▶ · 0:39
  12. Anthropic says Claude has a hidden “workspace” for silent reasoning▲ 460 · 199 comments▶ · 2:01
  13. Small, offline AI makes the case for edge-first deployments▲ 275 · 82 comments▶ · 2:53
  14. Mistral’s Leanstral 1.5 pushes open AI toward machine-checked proofs▲ 374 · 106 commentsin 5 episodes:▶ · 0:43▶ · 1:13▶ · 1:09▶ · 1:51▶ · 1:16
  15. Codex users spot a 516-token reasoning cliff▲ 371 · 151 commentsin 2 episodes:▶ · 1:08▶ · 1:58
  16. AMD inference gets cheaper per token, but HN wants the watt bill▲ 358 · 136 commentsin 5 episodes:▶ · 1:33▶ · 1:34▶ · 1:32▶ · 2:45▶ · 1:40
  17. Newer Claude models stumble on stricter tool schemas▲ 228 · 81 comments▶ · 1:56
  18. A hands-on guide to building a local LLM super-rig▲ 409 · 183 commentsin 5 episodes:▶ · 0:45▶ · 0:38▶ · 0:56▶ · 0:48▶ · 1:07
  19. Satire hits a nerve on ChatGPT in academia▲ 233 · 141 commentsin 3 episodes:▶ · 1:56▶ · 3:35▶ · 2:32
  20. AI and the end of math’s theorem economy▲ 284 · 122 commentsin 7 episodes:▶ · 2:06▶ · 0:36▶ · 0:36▶ · 0:44▶ · 0:41▶ · 2:18▶ · 0:41
  21. Turning code into images to cut Claude bills▲ 312 · 99 comments▶ · 1:34
  22. Zuckerberg says Meta’s AI-agent push hasn’t sped up as hoped▲ 248 · 232 comments▶ · 0:54
  23. Commerce lifts controls on Anthropic’s Fable and Mythos models▲ 975 · 690 commentsin 2 episodes:▶ · 0:33▶ · 3:04
  24. GitHub Copilot adds its first open-weight coding model▲ 417 · 184 comments▶ · 1:54
  25. Z.ai launches ZCode, a GLM-5.2 coding-agent harness▲ 509 · 355 comments▶ · 0:29
  26. Anthropic ships Claude Sonnet 5, and HN asks where it fits▲ 1264 · 783 commentsin 2 episodes:▶ · 0:33▶ · 1:41
  27. Qwen 3.6 27B makes local coding feel real — if your hardware can take it▲ 1188 · 757 commentsin 3 episodes:▶ · 1:56▶ · 0:27▶ · 2:07
  28. Anthropic launches Claude Science beta for lab workflows▲ 563 · 174 commentsin 2 episodes:▶ · 3:27▶ · 4:21
  29. Google debuts a faster, cheaper Nano Banana image model▲ 434 · 193 commentsin 2 episodes:▶ · 3:53▶ · 4:50
  30. Mistral posts Leanstral 1.5 for Lean theorem proving▲ 310 · 149 commentsin 2 episodes:▶ · 4:16▶ · 5:34
  31. Claude says Fable 5 is back — and users ask what they’re paying for▲ 407 · 418 comments▶ · 4:44
  32. LongCat-2.0: a trillion-parameter MoE trained on AI ASICs▲ 279 · 87 comments▶ · 5:10
  33. Cursor’s iOS app privacy toggle sparks backlash▲ 248 · 34 comments▶ · 5:58
  34. HackerRank’s AI resume scorer looks like a luck filter▲ 1027 · 433 comments▶ · 0:31
  35. OpenAI previews GPT-5.6 Sol—with faster agents and a gated rollout▲ 1134 · 743 commentsin 8 episodes:▶ · 0:32▶ · 0:29▶ · 0:32▶ · 0:35▶ · 0:35▶ · 0:33▶ · 0:28▶ · 0:32
  36. DeepSeek posts DSpark, a speculative decoding push for faster LLMs▲ 793 · 361 commentsin 8 episodes:▶ · 1:45▶ · 1:36▶ · 1:45▶ · 1:50▶ · 1:49▶ · 1:57▶ · 1:36▶ · 1:57
  37. Robin Williams, Good Will Hunting, and the case against AI slop▲ 401 · 225 commentsin 6 episodes:▶ · 3:07▶ · 2:39▶ · 2:53▶ · 2:54▶ · 2:57▶ · 3:31
  38. Asian startups pitch Mythos alternatives as U.S. export ban bites▲ 283 · 198 commentsin 6 episodes:▶ · 3:28▶ · 2:58▶ · 3:13▶ · 3:12▶ · 3:18▶ · 3:06
  39. Strix Halo gets a two-node RDMA playbook for local vLLM▲ 231 · 87 comments▶ · 3:15
  40. Open-weight LLMs may be closing the gap—but only on some benchmarks▲ 304 · 248 commentsin 2 episodes:▶ · 2:59▶ · 3:09
  41. Weave ships a router to pick cheaper coding models on the fly▲ 214 · 113 comments▶ · 3:21
  42. Open-weight AI models are putting frontier pricing under pressure▲ 201 · 186 comments▶ · 3:41
  43. A 744-billion-parameter open model squeezes onto local hardware▲ 615 · 305 commentsin 3 episodes:▶ · 0:33▶ · 0:33▶ · 1:50
  44. FUTO ships a new swipe-typing model trained on a million user swipes▲ 707 · 255 commentsin 2 episodes:▶ · 1:51▶ · 1:46
  45. Baidu’s Unlimited-OCR Tackles Long PDFs Without Blowing Up VRAM▲ 494 · 110 commentsin 2 episodes:▶ · 3:11▶ · 2:59
  46. Mistral ships OCR 4 with boxes, confidence scores, and self-hosting▲ 499 · 136 commentsin 2 episodes:▶ · 3:38▶ · 3:25
  47. Coding agents are entering the outer-loop era▲ 432 · 296 commentsin 2 episodes:▶ · 4:01▶ · 3:49
  48. AI’s cheap-token era may be ending▲ 332 · 420 commentsin 2 episodes:▶ · 4:28▶ · 4:15
  49. Anthropic brings Claude into Slack as a team agent▲ 268 · 182 commentsin 2 episodes:▶ · 4:55▶ · 4:39
  50. Swiss AI initiative releases Apertus Mini as an open, sovereign model set▲ 534 · 183 commentsin 5 episodes:▶ · 0:37▶ · 0:38▶ · 1:52▶ · 2:04▶ · 1:58
  51. GLM-5.2 gets close to Opus, but Hacker News wants better tests▲ 519 · 343 commentsin 3 episodes:▶ · 1:02▶ · 2:07▶ · 2:35
  52. Codex logging bug hammers local SSDs▲ 509 · 271 comments▶ · 1:27
  53. Open models are closing the gap, but Hacker News isn’t convinced it’s painless▲ 406 · 329 commentsin 4 episodes:▶ · 2:16▶ · 2:34▶ · 2:13▶ · 2:26
  54. Claude Code’s ‘extended thinking’ isn’t the real chain of thought▲ 326 · 225 commentsin 2 episodes:▶ · 2:40▶ · 3:02
  55. Moebius shrinks image inpainting to 226 million parameters▲ 321 · 83 commentsin 2 episodes:▶ · 3:00▶ · 3:25
  56. A 3B reasoning model claims frontier scores▲ 398 · 205 comments▶ · 3:27
  57. Sakana’s Fugu pitches model orchestration as the next AI product layer▲ 246 · 126 commentsin 2 episodes:▶ · 3:46▶ · 3:50
  58. Schmidhuber revisits Munich’s 1991 AI roots▲ 230 · 103 commentsin 2 episodes:▶ · 4:07▶ · 4:13
  59. Tiny Qwen model gets a big boost from fine tuning▲ 215 · 49 comments▶ · 4:43
  60. Open GLM model beats GPT-5.5 on hallucination rate, but HN isn’t buying the scaling claim▲ 584 · 292 commentsin 11 episodes:▶ · 0:32▶ · 0:35▶ · 0:35▶ · 0:38▶ · 0:37▶ · 1:34▶ · 0:31▶ · 2:56▶ · 3:32▶ · 3:12▶ · 4:06
  61. LLMs Have Entered Their Complicated Architecture Era▲ 212 · 76 commentsin 3 episodes:▶ · 2:19▶ · 1:54▶ · 2:32
  62. Cloudflare Rolls Out 60-Minute Temporary Accounts for AI Agent Deployments▲ 249 · 151 commentsin 3 episodes:▶ · 1:45▶ · 2:30▶ · 1:38
  63. Claude Rolls Out ID Verification, Sparking Privacy Backlash▲ 866 · 731 comments▶ · 2:08
  64. Midjourney Unveils Ambitious AI-Powered Medical Imaging Platform▲ 1377 · 877 commentsin 2 episodes:▶ · 0:29▶ · 1:22
  65. Local AI Models Are Catching Up, But Still Have Tradeoffs▲ 1596 · 607 commentsin 5 episodes:▶ · 0:35▶ · 0:39▶ · 0:37▶ · 0:34▶ · 0:37
  66. SpaceX to Acquire Coding AI Startup Cursor for $60 Billion▲ 1151 · 1702 commentsin 5 episodes:▶ · 1:35▶ · 1:31▶ · 1:29▶ · 2:18▶ · 1:20
  67. GLM-5.2 Sets New Standard for Open-Weights AI Models▲ 915 · 444 commentsin 5 episodes:▶ · 2:34▶ · 2:18▶ · 2:23▶ · 3:06▶ · 2:03
  68. ATProto Isn't Instance-Based Like Mastodon, and That's By Design▲ 536 · 315 commentsin 4 episodes:▶ · 3:16▶ · 2:43▶ · 2:50▶ · 3:28
  69. DeepSeek Adds Image Understanding to Its AI Chatbot▲ 498 · 204 commentsin 5 episodes:▶ · 3:40▶ · 3:53▶ · 3:36▶ · 3:48▶ · 2:20
  70. Local AI Models Like Qwen Aren't Just 'Worse Opus'—They're Different Tools▲ 492 · 253 commentsin 5 episodes:▶ · 3:58▶ · 4:14▶ · 3:54▶ · 4:26▶ · 2:35