frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Shopify moves back to Native from React Native

https://shopify.engineering/back-to-native
285•fnthawar2•1h ago•189 comments

Rust Is Tier-1 Language at Microsoft

https://rustfoundation.org/media/guest-post-rust-is-tier-1-language-at-microsoft/
214•mmastrac•2h ago•101 comments

Hitachi launches CO2 heat pump water heaters with solar-friendly tariff controls

https://www.pv-magazine.com/2026/09/07/hitachi-launches-co2-heat-pump-water-heaters-with-solar-fr...
179•thelastgallon•1d ago•144 comments

Amazon pilots ad services in ChatGPT

https://www.marketingdive.com/news/amazon-pilots-ad-services-in-chatgpt-what-marketers-need-to-kn...
76•thm•1h ago•69 comments

What algorithm did Windows XP use to choose your initial user picture?

https://devblogs.microsoft.com/oldnewthing/20260909-00/?p=112683
250•soheilpro•7h ago•118 comments

List of references on Sony websites to players "owning" their digital games

https://consumerrights.wiki/w/Sony_PlayStation_digital_game_ownership_lawsuit
191•haunter•3h ago•56 comments

>10x More Efficient Pretraining

https://magic.dev/blog/pretraining#
25•ronfriedhaber•1d ago•5 comments

More questions about whether researchers can trust OpenAI with unpublished math

https://mathstodon.xyz/@andreasthom/117240535270608201
14•pred_•9h ago•106 comments

To write non-fiction, draw the trunk, then the rest of the tree

https://devz.cl/posts/how-to-write/
41•DanielVZ•2d ago•7 comments

Stockfish 19

https://stockfishchess.org/blog/2026/stockfish-19/
174•atiedebee•2d ago•114 comments

iPhone Duo

https://www.apple.com/iphone-duo/
1330•thecosmicfrog•21h ago•2310 comments

Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra

https://cognition.com/blog/swe-2
9•seelos•35m ago•0 comments

Casablanca: How An Unproduced Play Marched into Movie History

https://www.thecollector.com/casablanca-unproduced-play-movie-history/
5•mdp2021•13m ago•0 comments

DeepSeek v4.1 Flash

https://twitter.com/deepseek_ai/status/2097930608790167907
721•Liwink•9h ago•378 comments

Serverless DTLS

https://proxylity.com/docs/listeners/dtls.html
3•mlhpdx•14m ago•1 comments

Show HN: What if the speed of light was 5 km/h?

https://rivendell.dmitrybrant.com/relativity/
517•dmitrybrant•14h ago•217 comments

Show HN: Filament – Fast data movement engine in Go

https://github.com/galaxy-io/filament
10•ikswolzok•2d ago•0 comments

The first drink-driving conviction may have happened in London

https://www.ianvisits.co.uk/articles/the-worlds-first-drink-driving-conviction-may-have-happened-...
8•beardyw•8h ago•9 comments

Who Dung It? (Turdle.fun)

https://turdle.fun/
11•nb_quant•2h ago•13 comments

What do Visa and Mastercard do? An intro to card networks

https://tautology.town/2026/06/01/card-networks.html
622•evakhoury•1d ago•377 comments

Discrete Dipole Approximation Code Ddscat 7.2 (2012)

https://arxiv.org/abs/1202.3424
7•doener•4d ago•1 comments

Liesegang Rings

https://chillphysicsenjoyer.substack.com/p/liesegang-rings
19•surprisetalk•1d ago•4 comments

Don't Let Anyone Take Away Your Big Box of Cables

https://blog.jim-nielsen.com/2026/hands-off-my-cables/
17•Brajeshwar•38m ago•10 comments

Show HN: Art – draw one stroke, let symmetry complete it

https://mrdee.in/mandala/
60•cyb0rg0•5d ago•26 comments

Tell HN: OpenAI keeps re-enabling the 'allow training' setting

331•jacquesm•2h ago•129 comments

Growing proof that autonomous cars save lives

https://spectrum.ieee.org/are-self-driving-cars-safe
418•bookofjoe•22h ago•727 comments

Larger Pacific Striped Octopus

https://en.wikipedia.org/wiki/Larger_Pacific_striped_octopus
108•olalonde•1d ago•53 comments

No Man's Sky Cosmos

https://www.nomanssky.com/cosmos-update/
429•Limb•1d ago•433 comments

GPT-6 Astra, looped transformers, and hidden reasoning

https://magazine.sebastianraschka.com/p/gpt-6-astra-looped-transformers-and
492•ModelForge•1d ago•158 comments

Bending Spoons buying Miro for $1.355B

https://investors.bendingspoons.com/newsroom/bending-spoons-agrees-to-acquire-miro
68•andygcook•2h ago•52 comments
Open in hackernews

>10x More Efficient Pretraining

https://magic.dev/blog/pretraining#
22•ronfriedhaber•1d ago

Comments

simonw•37m ago
> We match DeepSeek V4 Pro Base using ~50x fewer FLOPs – that’s around half of GPT3’s pretraining compute, or ~$0.5M on GB200.

If this holds up that's a really big deal.

wayfwdmachine•21m ago
Huge if true. As it were.
ismael_rr•12m ago
Super awesome. Wish they would release the paper about what they did to achieve this. I remember nous released the token superposition paper which improved pretraining FLOPs some, but not 50x: https://nousresearch.com/token-superposition. Wondering if they also found some cool tokenization strategiesa
monneyboi•7m ago
Imagine the sheer amount of power you could save by releasing the paper.