LeMario: Training a JEPA World Model on Super Mario Bros(benjamin-bai.com)

119 points by kevinjosethomas 4 months ago | 13 comments

HackEurope 2026: A short rant on AI and hackathons(duti.dev)

48 points by mindracer 2 months ago | 39 comments

Quadrillion-parameter transformers and how to train them(qlabs.sh)

60 points by frozenseven 10 days ago | 48 comments

Q Labs research post presenting a data-constrained scaling law with ensembling that prescribes far larger transformer models, up to quadrillion parameters, for fixed datasets.

NanoGPT Slowrun: 10x Data Efficiency with Infinite Compute(qlabs.sh)

153 points by sdpmas 7 months ago | 36 comments

J-space comparisons across open models(eliebak.com)

17 points by babelfish 4 months ago | 3 comments

Stateless MCP has recaptured my interest(simonwillison.net)

360 points by tosh 3 months ago | 204 comments

Simon Willison blog post on the stateless MCP 2.0 specification release and the mcp-explorer and datasette-mcp tools it inspired him to build.

A New Experiment Meta-Strategy(chillphysicsenjoyer.substack.com)

26 points by surprisetalk 1 month ago | 1 comments

A Substack post by an amateur physicist describing a new strategy for doing home experiments after reorganizing a messy lab to fit limited free evenings.

Auto-grading decade-old Hacker News discussions with hindsight(karpathy.bearblog.dev)

637 points by __rito__ 11 months ago | 261 comments

Andrej Karpathy blog post about using LLMs to grade decade-old Hacker News discussions and commenters' predictions with hindsight, presented as an "HN Time Capsule" site.

Recent AI model progress feels mostly like bullshit(lesswrong.com)

474 points by paulpauper 1 year ago | 375 comments

LessWrong linkpost essay arguing recent AI model improvements feel overstated, based on experience building an AI codebase security-scanning startup.

AI is dramatically accelerating AI research(planned-obsolescence.org)

333 points by yurivish 7 days ago | 300 comments

Ajeya Cotra's Substack post arguing a METR report shows AI is already roughly doubling AI capabilities progress at Anthropic, implying rapid compounding toward superintelligence.

The highest quality codebase(gricha.dev)

603 points by Gricha 11 months ago | 371 comments

Blog post by Greg Pstrucha describing an experiment in which Claude repeatedly reworked an app's codebase to make it the "highest quality", and what happened.

A new PNG spec(programmax.net)

610 points by bluedel 1 year ago | 553 comments

A blog post announcing the release of a new PNG specification after two decades of stalled progress and explaining why it matters.