Current AI models either use the personal pronoun "I" when referring to itself in its reasoning chain of thought, or "We". For example, R1 used "I", but R1-new uses "We". Will the best publicly available model at the end of 2025 use "We" more often than "I"?
For the purpose of this market, the best model at the end of 2025 is the model I think is the best, whose CoT is visible. Currently, Gemini 2.5 Pro, o3 and Claude Opus 4 are about as good but I would lean toward using o3 for the purpose of this market, if it were to close today (June 1st).
This market will close on December 1st, 2025, and resolve on January 1st, 2025.
Update 2025-09-18 (PST) (AI summary of creator comment): - Summarized CoT counts as visible CoT. If only a summarized chain-of-thought is available, it still qualifies as having visible CoT for this market.
Update 2025-12-31 (PST) (AI summary of creator comment): Creator has identified Opus 4.5 as the best model (as of this comment). The market will resolve based on whether Opus 4.5 uses "We" more often than "I" in its CoT.
1,000π Top traders
| # | Trader | Total profit |
|---|---|---|
| 1 | αΉ341 | |
| 2 | αΉ219 | |
| 3 | αΉ137 | |
| 4 | αΉ76 | |
| 5 | αΉ64 |
People are also trading
From the confessions paper by OpenAI, it looks like GPT-5 uses we (also in Fig. 18 with more details). Would this count? https://cdn.openai.com/pdf/6216f8bc-187b-4bbb-8932-ba7c40c5553d/confessions_paper.pdf
@LeonLinuxlu That's a good point. kinda cursed ambiguity. summarized CoT will count for the purpose of this market as visible CoT. If people wanna bet the alternative question (where summarized CoT don't count), lmk