What an unprocessed photo looks like

Growing up in “404 Not Found”: China's nuclear city in the Gobi Desert

Learn computer graphics from scratch and for free

You can make up HTML tags

Unity's Mono problem: Why your C# code runs slower than it should

MongoBleed Explained Simply

Software engineers should be a little bit cynical

PySDR: A Guide to SDR and DSP Using Python

As AI gobbles up chips, prices for devices may rise

62 years in the making: NYC's newest water tunnel nears the finish line

Researchers discover molecular difference in autistic brains

Show HN: My app just won best iOS Japanese learning tool of 2025 award (blog)

Spherical Cow

Why I Disappeared – My week with minimal internet in a remote island chain

Staying ahead of censors in 2025

How to complain (2024)

Slaughtering Competition Problems with Quantifier Elimination (2021)

Why I think Valve’s retiring the Steam Deck LCD

How small can a language model be while still doing something useful? I wanted to find out, and had some spare time over the holidays.Z80-μLM is a character-level language model with 2-bit quantized weights ({-2,-1,0,+1}) that runs on a Z80 with 64KB RAM. The entire thing: inference, weights, chat UI, it all fits in a 40KB .COM file that you can run in a CP&#x2F;M emulator and hopefully even real hardware!It won&#x27;t write your emails, but it can be trained to play a stripped down version of 20 Questions, and is sometimes able to maintain the illusion of having simple but terse conversations with a distinct personality.--The extreme constraints nerd-sniped me and forced interesting trade-offs: trigram hashing (typo-tolerant, loses word order), 16-bit integer math, and some careful massaging of the training data meant I could keep the examples &#x27;interesting&#x27;.The key was quantization-aware training that accurately models the inference code limitations. The training loop runs both float and integer-quantized forward passes in parallel, scoring the model on how well its knowledge survives quantization. The weights are progressively pushed toward the 2-bit grid using straight-through estimators, with overflow penalties matching the Z80&#x27;s 16-bit accumulator limits. By the end of training, the model has already adapted to its constraints, so no post-hoc quantization collapse.Eventually I ended up spending a few dollars on Claude API to generate 20 questions data (see examples&#x2F;guess&#x2F;GUESS.COM), I hope Anthropic won&#x27;t send me a C&amp;D for distilling their model against the ToS ;PBut anyway, happy code-golf season everybody :)

Show HN: Z80-μLM, a 'Conversational AI' That Fits in 40KB

Fast CVVDP implementation in C