DigestAI news desk

Cut through the AI noise.

Research3 min read

Developer trains 12.5M-parameter AI to pick Pokémon starter on single GPU

A developer named stmonty trained a small AI model with just 12.5 million parameters on a single RTX 3080 Ti to play Pokémon Red. The model, built using a world model approach inspired by research from Yann LeCun’s AMI Labs, learns button presses by analyzing screen outputs rather than predicting full frames. It compresses visual data into 192 numbers and uses reward-free training initially,…

1 source

Key points

  • Developer stmonty trained a 12.5M-parameter AI on a single RTX 3080 Ti to play *Pokémon Red*,
  • Model uses compressed screen data (192 numbers) and reward-free training to learn button presses,
  • After fine-tuning, AI picked Squirtle starter in 52% of 100 runs, outperforming random presses

The model required 42,382 grayscale frames across over 1,000 short runs to learn basic mechanics. After fine-tuning, it successfully selected Squirtle as a starter in 52 out of 100 runs, outperforming random presses and an untrained model. Stmonty notes compounding errors in predictions as a key challenge. The project, open-sourced on GitHub as lePokeRed, demonstrates how lightweight models can tackle niche tasks on consumer hardware. While other approaches have beaten Pokémon Red using larger models, this work highlights the potential for hobbyist experimentation.

Full story from Tom's Hardware · by Shane DowningOpen source ↗

Developer trains a small AI on a single RTX 3080 Ti gaming GPU to 'play' Pokémon Red

Tom's Hardware · 30 September 2026

Loading the full article…

This text was published by Tom's Hardware and written by Shane Downing. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗

Topics · follow one to build your own front page
AMI LabsLeWorldModelJEPA modelstmontyYann LeCun

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.

Comments

via GitHub Discussions

More in Research

All →

Related stories