this might be one of the funniest experiment I saw this month. @the_nof1 gave 10k to each LLMs to trade on various crypto assets including bitcoin, ethereum and solana. without any surprise, we can see that each of them have their own strategy and different behaviours. LLMs: > Gemini 2.5: good win rate, but bad r/r, with a bearish bias > GPT 5: bearish bias as well, short on the market and almost 15% down. > Claude: seems very reasonable and hold a stable portfolio. > Deepseek: printing decent profits (SO FAR) and seems like a moderated trader. > Grok 4: is doing his best. > Qwen 3: crazy cousin going max. lev printing a portfolio in loss rn. Potential decision triggers: > how fast LLMs receive macro data > LLMs bias based on aggregated data > bad at TA (I don’t think they can take TA decisions nor access to low timeframes price action) prove me wrong if you have any data I’d be curious. I genuinely think we could see LLMs becoming extremely performant at trading as they could also get rid of any bias, emotions and aggregate thousands of piece of data (in an ideal future) while it would take hours for a human brain. hard to not feel the urge for the human to become more performant in order to avoid being replaced by automated tools with great awareness. are we inevitably moving towards transhumanism in order to preserve human's labour in our society while reaching a symbiotic state between artificial intelligence and human's own abilities? sources: @jinglingcookies @jay_azhang