A hobbyist built a 12.5 million‑parameter world model on a single RTX 3080 Ti to pick a starter in Pokémon Red.
NoteOossaPublished by Oossa: 1 min read
A developer known as “stmonty” trained a small AI called a world model on one RTX 3080 Ti GPU. The model has about 12.5 million parameters and learns to choose a starter Pokémon from a saved game. The work follows research by Yann LeCun on world models and was posted on the developer’s blog on Sept. 20, 2026. It runs locally on a consumer‑grade graphics card.
It shows that modest hardware can run AI that understands game states, lowering the barrier for hobbyist AI experiments.
Last updated: ·Markdown
Models
Deepseek announced TileLang and related libraries as open‑source software for Huawei's Ascend 950 chips, aiming to simplify AI development in China.
Models
Lokutor’s open‑source Oído model achieves lower error rates than Whisper‑tiny while running on a cheap ESP32‑S3 board.
Models
About 100 sites get paid for content used in AI Overviews, AI Mode and Gemini, but most earn under $1,000 and don’t know how payouts are calculated.
Oossa · Newsletter
The week in AI, explained
Every Monday: the stories worth knowing, in plain language. Free, no spam.
