frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: Andrej Karpathy's microgpt.py to C99 microgpt.c – 4,600x faster

https://github.com/enjector/microgpt-c
34•Ajay__soni•1h ago
Andrej Karpathy showed us the GPT algorithm. I wanted to see the hardware limit.

The Punchline: I made it go 4,600x faster in pure C code, no dependencies and using a compiler with SIMD auto-vectorisation!!!

Andrej recently released microgpt.py - a brilliant, atomic look at the core of a GPT. As a low-latency developer, I couldn't resist seeing how fast it could go when you get closer to the metal.

So just for funzies, I spent a few hours building microgpt-c, a zero-dependency and pure C99 implementation featuring:

- 4,600x Faster training vs the Python reference (Tested on MacBook Pro M2 Max). On Windows, it is 2,300x faster. - SIMD Auto-vectorisation for high-speed matrix operations. - INT8 Quantisation (reducing weight storage by ~8x). Training is slightly slower, but the storage reduction is significant.

- Zero Dependencies - just pure logic.

The amalgamation image below is just for fun (and to show off the density!), but the GitHub repo contains the fully commented, structured code for anyone who wants to play with on-device AI.

I have started to build something useful, like a simple C code static analyser - I will do a follow-up post.

Everything else is just efficiency... but efficiency is where the magic happens

Comments

idiotsecant•12m ago
I get that the LLM wrote the code but can you at least write the description? I am getting second hand halitosis of the soul reading all these obvious slop project posts. Use your human mind to write human words to tell us about why this is important to you.

Ask HN: What is something someone else did that made your day better?

1•blahaj•24s ago•0 comments

Show HN: OpenEntropy – 47 hardware entropy sources from your computer's physics

https://github.com/amenti-labs/openentropy
1•amentiflow•42s ago•0 comments

Shard – A Distributed P2P AI Network for Shared Inference

https://github.com/TrentPierce/Shard
1•tpierce89•1m ago•1 comments

Earn $50 PER REFERRAL and $2 PER CLICK

https://hunnypack.com/
1•bokeke1•4m ago•0 comments

A fluid can store solar energy and then release it as heat months later

https://arstechnica.com/science/2026/02/dna-inspired-molecule-breaks-records-for-storing-solar-heat/
1•pseudolus•5m ago•0 comments

Could an Electronic Real-Time Coach Help Ski Jumpers Leap Farther?

https://www.nytimes.com/2026/02/15/science/olympics-technology-ski-jump.html
1•bookofjoe•11m ago•1 comments

The End of the Office

https://blog.andrewyang.com/p/the-end-of-the-office
2•cebert•14m ago•0 comments

Meta patented an AI that lets you keep posting from beyond the grave

https://www.businessinsider.com/meta-granted-patent-for-ai-llm-bot-dead-paused-accounts-2026-2
1•johnhamlin•15m ago•0 comments

Show HN: Task Automation Analysis of Labor Statistics and O*Net Jobs Data

1•Falimonda•17m ago•0 comments

I don't think AI performance will plateau

https://honnibal.dev/blog/ai-bubble
1•syllogism•19m ago•2 comments

Racket Syntax: The Great, the Good and the Back-to-the-Drawing-Board (2024) [video]

https://www.youtube.com/watch?v=ZtTqRH1uwu4
1•so-cal-schemer•20m ago•1 comments

MacKenzie Scott's $26B Sugar Pile

https://garryslist.org/posts/mackenzie-scott-s-26-billion-sugar-pile
2•gmays•21m ago•0 comments

Game developers and pixel artists are losing their jobs

https://www.sprite-ai.art
1•tjco•22m ago•2 comments

What would a "permissions-first ORM" look like? Looking for spec feedback

https://typescript-superapp.bunnytech.app/docs
1•iosifnicolae2•22m ago•2 comments

Instagram boss says 16 hours of daily use is 'problematic' not addiction

https://www.bbc.com/news/articles/cn71mgmzljlo
1•pseudolus•24m ago•0 comments

Earthquake Magnitude Scale

https://www.mtu.edu/geo/community/seismology/learn/earthquake-measure/magnitude/
1•teleforce•27m ago•0 comments

India's 'AI Impact Summit' Promises Little More Than Spectacle

https://internetfreedom.in/indias-ai-impact-summit-promises-little-more-than-spectacle/
1•akbarnama•28m ago•0 comments

AI Writes Code in Seconds. Why Do Your Tests Take Minutes?

http://stumpy.ai/blog/your-ai-writes-code-in-seconds
1•bluesnowmonkey•29m ago•2 comments

Robert Duvall, Oscar-winning actor and 'Godfather' mainstay, dead at 95

https://www.cnbc.com/2026/02/16/robert-duvall-dies-at-95.html
2•pseudolus•32m ago•1 comments

Rare Pokemon card sets record with $16.5M sale

https://www.japantimes.co.jp/life/2026/02/16/digital/pokemon-card-sale-most-expensive-pikachu-ill...
1•anigbrowl•33m ago•0 comments

Beating GPT-2 for less than $100 – Andrej Karpathy

https://github.com/karpathy/nanochat/discussions/481
2•logicprog•38m ago•0 comments

Show HN: Bulwark – Open-source governance layer for AI agents (Rust, MCP-native)

https://github.com/bpolania/bulwark
1•bpolania•40m ago•2 comments

Ask HN: Best roles in tech where I can be in meetings mostly?

2•general_reveal•42m ago•2 comments

Vulnerabilities in cloud-based password managers [pdf]

https://eprint.iacr.org/2026/058.pdf
3•leobdkr•44m ago•2 comments

Ask HN: Which password manager do you use / would you recommend?

3•unodonut•46m ago•10 comments

Linux CVE Assignment Process

http://www.kroah.com/log/blog/2026/02/16/linux-cve-assignment-process/
2•LorenDB•48m ago•1 comments

Lack of measurement invariance in mental health across intelligence levels

https://www.sciencedirect.com/science/article/abs/pii/S0160289625000662
1•i7l•49m ago•0 comments

Show HN: Krea iPad – real-time editing model with Apple Pencil input

https://twitter.com/venturetwins/status/2023107207500566675
1•dvrp•49m ago•0 comments

Dark web agent spotted bedroom wall clue to rescue girl from abuse

https://www.bbc.com/news/articles/cx2gn239exlo
75•colinprince•52m ago•28 comments

Meta: Messenger.com is no longer available for messaging

https://www.facebook.com/help/messenger-app/804132271957789
3•ddxv•53m ago•1 comments