The self-improving brain.

I just read DeepMind's new paper, “From AGI to ASI” (Genewein et al., 2026; arXiv:2606.12683). It's an unusually readable one, and it got me thinking about what we're building at FinWhale.

The paper's argument is that AI doesn't become superhuman in a single moment — it climbs there gradually. And one of the paths up is systems that improve themselves: they learn from what they do and get sharper each cycle. That framing lined up exactly with the bet we've been making.

FinWhale is two things.

One is a layer of market data you can actually trust — reliable data infrastructure, built with point-in-time discipline in mind.

The other, on top of it, is a smart, self-improving brain that assists the trader through the everyday routine — research, idea generation, backtesting and hypothesis-testing. It learns how you work, keeps watch on the market, and proactively tells you what's worth looking at next.

Here's the part I find cool.

Say you notice a hunch — some pattern in software names this earnings season. Normally that just lives in a notebook, in Notion, in OneNote, and quietly dies there.

With FinWhale you can test it properly against years of history. If the edge is real, it gets remembered. The system watches for the next time that setup appears, and weeks later it comes back to you on its own:

Less a chatbot. More a brain that keeps score.

So it's less a chatbot you re-explain yourself to every morning, and more a brain that accumulates what you've learned, keeps score on it, and gets sharper as it goes.

FinWhale comes in two forms — a web app, and a version that lives right inside the AI tools traders already use.

Building in public from here.