Claude 3 Opus knows who I am. It knows my favorite programming languages, my text editor, and when I ask questions like "What 20 approaches might [Mira] consider for Problem X?" a full 15-20% of its answers are uncommon techniques that are actually in my idea backlog. ("Dependently-typed genetic programming")

I'll test the model and see if it knows who I am, and resolve:

• YES if I'm convinced it knows who I am

• 50% if I can't tell

• NO if I'm convinced it doesn't know

• N/A if I can't access Llama 3 405B by end of year

Claude 3 Opus is the only model I believe knows who I am; GPT-4, Llama 70B, Gemini 1.5 Pro, smaller Claude 3 models or Claude 3.5 Sonnet, none of them know who I am.

It doesn't need to link me to @Mira on Manifold Markets.

I won't bet in this market.

I think it's unlikely this will resolve 50%.

I may not reveal the prompts I used if they would identify me.

I might wait for it to appear on Groq to test it.