Abdullah Shareef
9 posts

Abdullah Shareef
@itsabdullahs
Founder @ ScalarFX, ex-Quant Trader @ Susquehanna

I would like to purchase this @ROLEX






AOC: “There’s a certain level of wealth and accumulation that is unearned. You can’t earn a billion dollars. You just can’t earn that. You can get market power, you can break rules, you can abuse labor laws, you can pay people less than what they’re worth, but you can’t earn that”






What happens when you let an LLM autonomously improve its own prediction market strategy for 9 days straight? It learns to give up. Me and @ahall_research worked on a LLM pipeline on Polymarket (Bitcoin, NYC weather, miami weather) and every 4 hours it made predictions. Every night a separate AI reviewed performance, picked one thing to change and committed code. No human in the loop. Timeline: - Day 1: It added confidence prompts - "don't hedge, commit to your prediction" - Day 2: It turned down the shrinkage parameter that pulls predictions towards market consensus - performance got worse - Day 4: It reversed course and turned shrinkage back - Day 6: It tried a clever prompt trick for weather contracts (also made things worse and got autoreverted) - Day 7-9: It stopped changing anything By the end, the squared error was down 54%. This sounds impressive until you realize that the improvement came almost entirely from the system learning to deviate less from market prices. The self improvement loop worked perfectly as engineering but the problem is that the underlying system doesn't have an information edge on any of these contracts since it's reading the same news the market already priced in. To actually beat the market, the system would likely need faster or better information or contrarian accuracy (i.e., correct predictions specifically in the cases where the market is wrong).



