Osamu Dazai

AI Researcher & PhD | prop trading quant ｜ex-tech research scientist / machine learning engineer

[Osamu Dazai @ijstephdz]

Tmr I am gonna share more on the technical side of ai modeling for trading.

[Osamu Dazai @ijstephdz]

Recently, I’ve been thinking about AI agents, world models, etc.

Share some basic thoughts.

Specifically, I’m thinking about trading. However, I don’t think that an AI agent (LLM-based) could come up with good strategies as a single component. HFT? No way.

[Osamu Dazai @ijstephdz]

Frankly, I’m on the fence about joining FAIR or the Gemini team. Nobody would pass up the chance to put their name on Gemini, GPT, or LLaMA’s technical report. Impact and attention are money too. Joining a trading firm means you’d need to cut off your publication flow.

[Osamu Dazai @ijstephdz]

There’s FOMO in the quant industry. Big firms are recruiting LLM experts. But it’s awkward that they’re not that attractive to top AI researchers (as far as I know, e.g., PhDs from top U.S. universities). #Meta #OpenAI #Anthropic are paying more than #JaneStreet #HRT #Citadel

[Osamu Dazai @ijstephdz]

HFT? No way. If you basically knew how HFT works, you’d see why. MFT still needs a lot of improvement. Especially, LLMs are too bad at time series modeling. EDA needs sanity checks too. But I do think multi-agent systems have the potential to mimic a good trader’s behavior.