Latest
22
Nov
Validation debt
AI is driving the cost of producing a scientific result toward zero. The cost of checking one isn't falling at all. Every result we make faster than we can validate it is a loan against the bench, and I call it validation debt.
4 min read
17
Nov
Benchmarks can only grade known science
A benchmark needs a known answer before the test runs, so it can only grade science we already know. That makes every science benchmark a surrogate endpoint for discovery. Use a leaderboard to pick what to test at the bench, never as the result.
3 min read
09
Jun
I'm Back
My one project, ten times as awesome, turned out to be an ecosystem: data pipelines, AI agents, and Heureka Bench, a local lab notebook with an agent built in. It's called Heureka Labs, and the writing is back.
3 min read
27
Dec
The limits of AI
Asking ChatGPT for a realistic Western blot shows three limits of AI: models are censored, they're improving fast, and they don't yet solve real problems for scientists. Like any new tool, we find its limits to learn what it's good for.
4 min read
17
Dec
A lot can happen in a year
Five lessons from a year of weekly writing: build a system, experiment, increase your chance of luck, build and remix, and focus. Heureka Labs is about how technology can amplify creativity, particularly in science.
3 min read
18
Dec
Flipping Lab Meetings
Flipped lab meetings move presentations to recorded video, watched on your own time, and save meeting time for discussion. People can pause, ask questions in writing, and let ideas simmer, so everyone gets a voice.
3 min read
04
Dec
Enhancing Creativity With AI
ChatGPT wrote this whole post, from the outline to the poems, as a demonstration of what it can do. It isn't really about chat. Language generation this good will change how we write and create.
4 min read
21
Aug
Synthesizing
A year of weekly essays circled one question: how do you come up with a good idea? Next I'm synthesizing them into a course, built on the platform I made for teaching data science, because teaching is how you master something.
2 min read
07
Aug
Wait
A minimum viable mindset says to get your ideas out into the world early. But when an idea isn't ready yet, have the confidence and foresight to hold it back.
31
Jul
Minimum Viable Mindset
Science needs minimum viable discoveries: atomic, right enough to build on, and shared early. A minimum viable mindset is an experimental one. Make the best small thing you can, get feedback, and design the next experiment.
3 min read