Content
@
https://warpcast.com/~/channel/gray
0 reply
0 recast
0 reaction
✳️ dcposch on daimo
@dcposch.eth
I gave the leading AIs the Thiel question. First thing first, Claude absolutely smoked ChatGPT. ClosedAI "thought for 5 seconds" (hidden proprietary chain of thought) to come up with the exact opposite of a good answer--a belief that is basic, widely held, and wrong. It's all luck bro.
5 replies
15 recasts
70 reactions
✳️ dcposch on daimo
@dcposch.eth
Gemini and Grok gave self-referential answers about AI. Gemini is so cowed by its safety RLHF that it answered defensively! "My unpopular belief is that AI... is not so bad actually! Please no negative rewards!"
1 reply
0 recast
2 reactions
✳️ dcposch on daimo
@dcposch.eth
But I saved the best for last. Look at this absolutely beautiful answer. ExaAI x DeepSeek. Best answer = only open source model on the list. Exa built this in a few days as a hackathon project. 2025 is the year of open models and permissionless invention. Strap in. https://demo.exa.ai/deepseekchat
2 replies
1 recast
5 reactions
Mark Fishman
@markfishman
This is the best final output but it's strange to see the search results + citations for a question that's supposed to be "personal". Obviously none of these models create truly original output, just noting a weird contrast
1 reply
0 recast
1 reaction
Johnson
@johnson
Interesting how majority of them are self referential like you said. This is also from deepseek but the webUI deepseek host officially.
1 reply
0 recast
0 reaction
