Writing · WeChat “Vega二向箔”
Would you hand your brokerage account to an AI agent trading 24/7? Following Robinhood's experiments, three layers of logic: from tool to agent, where alpha comes from when everyone uses AI, and why so many people resent AI — a lack of skin in the game.
Read more →Notes on Caixin's cover story on the token economy: as AI moves from answering a question to finishing a whole job, enterprise bills, model pricing, industry profits and the battle for the office are all shifting. Growth and profit must be counted separately; ordinary people's opportunity lies in getting things done.
Read more →95% of enterprise AI projects fail — what do the other 5% do? Unpacking the playbook of OpenAI FDE lead Colin Jarvis: decide whether you earn service fees or product revenue, eat pain and excrete product, avoid premature generalisation, and the Klarna and Morgan Stanley cases — the hard part is the data layer.
Read more →An LLM that can write factor formulas is not a usable quant research agent. Factor-Agent wires PandaAI's data, factor computation, backtesting and QuantSkills into a verifiable, traceable, iterative research loop: the LLM only proposes hypotheses; evidence from backtests, out-of-sample tests and overfitting gates decides what gets promoted.
Read more →