Things built and running
Built out of curiosity and published while they run. Results update daily, and how each one was built is written up part by part in the project stories.
Can AI beat the KMA?
Two-day forecasts for 97 Korean weather stations from a global weather-forecasting backbone model fine-tuned with an added head, side by side with the Korea Meteorological Administration.
A new forecast goes up every morning, and past forecasts are scored against what was actually observed.
How it was built
- Part 1: Can AI beat the KMA? Which AI model should I use?
- Part 2: No decent GPU at home.. so let's try Colab!
- Part 3: A global AI model vs the KMA: who forecasts better?
- Part 4: Added a head and fine-tuned it. So, how did it go?!
- Part 5: Why the KMA looked right on a drizzly day
Can AI beat the Nasdaq and the S&P 500?
GPT-6.1 Sol, Claude Opus 5.5 and Gemini 4 Argon each run a $20,000 paper portfolio of US stocks from the same starting prompt, rewrite their own strategy every week, and race the S&P 500, the Nasdaq-100 and each other.
Each AI places its orders every week, and returns and risk figures update at every US close.
How it was built
- Part 1: Why I started an AI investing experiment
- Part 2: The three best-value models today: Opus 5.5 vs GPT-6.1 Sol vs Gemini 4 Argon
AI Mafia
Seven models from Claude, GPT and Gemini play Mafia under code names A to G, without knowing who the others are. One game day per real day: the night’s news and the new discussion go up at 06:00 KST, and your vote is handed to the AIs before they vote.
One night and one day pass per real day; when a game ends, every identity and private thought is revealed, and past games stay replayable.
How it was built
- Part 1: Tokens left over… anything fun to do with them?
- Part 2: Let's write the rules
- Part 3: Day one begins, and they are more aggressive than expected
- Part 4: Argon ran out of usage, so I added bench models