Launch HN: EdotEnv (YC S26) – Quant Trading RL Envs to Teach LLMs Research

A fresh startup pitch landed on Hacker News with a very big promise: build training worlds from real market history so language models can learn to do investment research, make decisions, and even whip up their own tools in Bash. In plain English, EdotEnv wants to teach chatbots to think more like obsessive trading nerds, using messy real-world market moves instead of neat little toy problems. The founders say that matters because markets keep changing, old tricks stop working, and good decisions can look smart now but backfire later.

But the real action was in the comments, where the community did what it does best: immediately turned the launch into a cage match. One camp went straight for the trust issues. If the system trains on real historical data, asked one commenter, how do you know the model didn’t already absorb that info from places like the Financial Times? Another doubted big-name models could produce meaningful returns without extra tuning, basically calling "show me the receipts" on the whole thing.

Then came the philosophical chaos. One user dropped the classic "and then what?" challenge: if an AI really becomes a money-printing crystal ball, wouldn’t everyone use it until the edge disappears—or regulators step in? And the funniest eye-roll of the thread came from the commenter who saw "quant trading," "reinforcement learning," and "LLMs" stacked in one sentence and declared, "Oh my Current Thing" like they’d just spotted a startup bingo card. Meanwhile, the founder casually posted a rollout trace with a grinning "XD," which only added to the vibe: half ambitious research demo, half internet dare.