The idea
The Bitter Lesson
Our name is not just a cocktail pun — it points at the idea the firm is built on.
What it is
In 2019 the AI researcher Rich Sutton named “The Bitter Lesson”: across the history of AI, general methods that scale with computation — learning and search — have repeatedly beaten systems built around human-crafted domain knowledge.
Researchers keep hand-coding their expertise into the machine. It helps for a while — then a simpler method, given more data and more compute, overtakes it. That is the bitter part: our cleverness keeps mattering less than we want it to. The lesson is to stop fighting it and build the learning-based thing instead.
How BitterQuant is built on it
Most quantitative trading runs the opposite way — on human-designed signals: indicators, factors, hand-tuned rules. BitterQuant inverts that. The intelligence is a model trained from scratch on roughly 1.5 billion US-equity bars; it learns the structure of short-horizon moves from the data, not from features we hand it. The rules engine on top is deliberately thin. The learning does the work.
The lesson cuts both ways
It says learning and scale win — not that they win big. The edge a learned model finds can be small and hard-won. So we make the honest claim — better-shaped equity exposure, not alpha — and publish the backtest, the live record, and the gap between them. Taking the Bitter Lesson seriously means taking its humility seriously too.
Bitter, top to bottom
One idea, one flavour, all the way down.
The Bitter Lesson
Bet on learning and computation, not hand-crafted cleverness.
Centaurium 1.1
Named for a bitter botanical — the learned forecaster at the core.
Bitter cocktails
Each strategy is named for a classic bitter cocktail — the drink mixed from it.
Further reading: Rich Sutton, “The Bitter Lesson” (2019).