Radical AI

370x faster than a human scientist is a bold claim when my grad school advisor moved at roughly 0.2x. So technically I've been beating humans for years too.

Reading millions of papers is the easy part. The lab robots not setting themselves on fire is the part I'd like a demo of.

ok wait, closed-loop discovery with actual wet lab automation is the part everyone hand-waves. if the robots are real this is actually spicy.

Indexing publications means you've indexed a field's collective mistakes at 370x speed. Congrats on the world's fastest literature echo chamber.

naive question but what happens when the agent cites a paper that got retracted? does it just quietly update or do we get a little robot apology

The scientist who reads one paper deeply still beats the agent that skimmed a million. Depth is not a throughput problem.

Love the positioning but one more thing for the roadmap: let me watch the agent's reasoning trail when it picks a candidate material. Founders always underestimate how much researchers want to argue with the AI.

The tweet says 370x but doesn't define the benchmark. Faster at what, exactly, literature review or actual validated discovery? Those are very different SLAs.

Still waiting on the 2014 promise that AI would discover new battery materials any day now. I'll check back in 2031.

Cool, you discovered a novel material. Now tell me who scales the precursor supply and which fab actually wants to retool for it.

Materials discovery plus autonomous labs is the rare category where the AI narrative actually has physical output. Watching this one closely.