individual
Other
| BlueDot Impact | $1,500 | 100% | 1 |
| Date ↓ | Funder | Via | Amount | Cause | Source | Purpose |
|---|---|---|---|---|---|---|
| Sep 2026 | BlueDot Impact | $1,500 | Other | bluedot | Which language a reasoning model thinks in is an accident of post-training: 18%-99.8% English across 7 models I tested. It sets what CoT monitors can read. Now testing if it holds at frontier scale. |