Post

Conversation

David Watson 🥑
Post your reply

Translated from Chinese
AI's finally willing to step out of the lab and do the dirty work—that's a good thing. Running benchmarks on the computer all day is fucking useless; the real skill is figuring out how to use a pipette and actually making that protein binder work. If it can nail the closed loop,
Translated from Chinese
Only when the agent is hooked up to the experimental robot can the paper-based closed loop be considered truly implemented. What I'd really like to see is whether, when it encounters a pipe without a label, it'll obediently stop and ask a human.
Google just put agents in the actual wet lab, so now the CI pipeline can fail at 2am over something nobody can reproduce locally. At least when my builds break, they're not also setting a reagent on fire.
I'd love a worked example of one failed round: what did the measurements rule out, and how did that change the next experiment?
This closed-loop lab workflow could really speed up protein binder research for cancer targets.
The aptamer experiment has a lesson that applies well beyond biology. Raw binding increased in one round, but much of it came from molecules sticking to things they weren't supposed to. The number looked better while target specificity got worse. A good question for any team
Is there a public run-level trace that includes failed experiments as well as successful ones? I’d like to separate protocol-execution errors from biologically negative results when evaluating how the loop improves.
I actually achieved two-host byte-identical reproduction across Intel and AMD using a staged reference backend — artifacts matched. I've also implemented an event-sourced agent substrate with hash-chained history, incremental verification, checkpoints, and deterministic state
Square profile picture
The next generation Starlink V5 has a smaller form factor and lightweight design with greater power efficiency. With speeds up to 375+ Mbps, Starlink V5 delivers reliable home internet for streaming, video calling, gaming and more. Available in select areas.
AI agents combined with lab robotics could dramatically accelerate cancer research. Discovering highly effective protein binders through faster design-and-test cycles is an exciting step toward the future of biotechnology!
How are failed runs and ambiguous readouts represented back to the agent? Keeping those alongside successes seems essential for understanding what the loop has actually learned.