HN Atlas
← All topics

Machine learning research and practice

Neural networks, transformers, learning methods, model training and evaluation, and practical machine-learning research.

41,140 stories · 162 in 7 days · Updated Nov 21, 2026

01 / CATCH UP

The conversation now

Popular recent stories, shown newest first.

A few hours ago Google DeepMind created a math horizon repo↗ github.com · HN · 31 comments · 44 points · Nov 20, 2026
Language Models Can Rewrite Their Own Weights↗ arxiv.org · HN · 120 comments · 411 points · Nov 20, 2026
Claude was trained on protein sequences at scale↗ xinmingtu.cn · HN · 76 comments · 181 points · Nov 18, 2026
Weight-Space Programming builds neural networks like a 1990s researcher↗ notesbylex.com · HN · 13 comments · 122 points · Nov 17, 2026
What is the best harness for each model?↗ runta.com · HN · 134 comments · 421 points · Nov 17, 2026

02 / EXPLORE THROUGH TIME

Timeline

Start with the latest. Scroll back through time, or zoom in for more.

Year by yearUp to 5 stories per year
20266,682 stories
20256,481 stories
20246,545 stories
20236,879 stories
20224,239 stories
20214,094 stories
20206,220 stories

The pace of conversation

Posts per month

2020-01: 516 posts2020-02: 506 posts2020-03: 456 posts2020-04: 568 posts2020-05: 590 posts2020-06: 613 posts2020-07: 583 posts2020-08: 532 posts2020-09: 461 posts2020-10: 470 posts2020-11: 475 posts2020-12: 450 posts2021-01: 338 posts2021-02: 343 posts2021-03: 381 posts2021-04: 345 posts2021-05: 339 posts2021-06: 363 posts2021-07: 340 posts2021-08: 359 posts2021-09: 334 posts2021-10: 341 posts2021-11: 294 posts2021-12: 317 posts2022-01: 312 posts2022-02: 269 posts2022-03: 342 posts2022-04: 363 posts2022-05: 420 posts2022-06: 377 posts2022-07: 364 posts2022-08: 332 posts2022-09: 341 posts2022-10: 356 posts2022-11: 355 posts2022-12: 408 posts2023-01: 449 posts2023-02: 489 posts2023-03: 656 posts2023-04: 715 posts2023-05: 606 posts2023-06: 570 posts2023-07: 518 posts2023-08: 538 posts2023-09: 538 posts2023-10: 628 posts2023-11: 568 posts2023-12: 604 posts2024-01: 553 posts2024-02: 520 posts2024-03: 621 posts2024-04: 625 posts2024-05: 619 posts2024-06: 522 posts2024-07: 557 posts2024-08: 481 posts2024-09: 443 posts2024-10: 563 posts2024-11: 559 posts2024-12: 482 posts2025-01: 580 posts2025-02: 516 posts2025-03: 528 posts2025-04: 542 posts2025-05: 544 posts2025-06: 484 posts2025-07: 537 posts2025-08: 596 posts2025-09: 486 posts2025-10: 551 posts2025-11: 558 posts2025-12: 559 posts2026-01: 587 posts2026-02: 633 posts2026-03: 637 posts2026-04: 582 posts2026-05: 563 posts2026-06: 623 posts2026-07: 614 posts2026-08: 563 posts2026-09: 700 posts2026-10: 730 posts2026-11: 450 posts
2020-012026-11

Posts per month · first and latest months may be partial.

04 / GO DEEPER

Follow your curiosity

Find a particular story, revisit a year, or see what resonated.

Highest rated · the last 30 days

Astra can read and write weights to improve Qwen↗ rikiyat.github.io · HN · 902 comments · 2971 points · Nov 02, 2026
Kimi K3.1 Pro and K3.1 Pro Highspeed↗ kimi.ai · HN · 682 comments · 1096 points · Oct 23, 2026
Sebastian Raschka's from-scratch video course on building reasoning models↗ magazine.sebastianraschka.com · HN · 116 comments · 857 points · Oct 26, 2026
GPT-6.1 Terra↗ openai.com · HN · 428 comments · 638 points · Oct 26, 2026
Dessa: Conversation Model trained only on speech↗ block.xyz · HN · 199 comments · 537 points · Oct 23, 2026
E-means: Outperforming k-means in one-tenth the epochs↗ e-means.github.io · HN · 93 comments · 522 points · Nov 09, 2026
Anthropic blocks users for anything touching training larger models↗ twitter.com · HN · 273 comments · 497 points · Oct 31, 2026
What is the best harness for each model?↗ runta.com · HN · 134 comments · 421 points · Nov 17, 2026
Language Models Can Rewrite Their Own Weights↗ arxiv.org · HN · 120 comments · 411 points · Nov 20, 2026
Cerebras Deploys Trillion-Parameter AI Models↗ cerebras.ai · HN · 142 comments · 368 points · Oct 25, 2026

10 stories shown.

Help shape Hacker Atlas

Useful, confusing, or missing something? I’d love to hear.

Your feedback is private. Your email is only used to reply, never to sign you up for a newsletter.