Anthropic built a chatbot that hallucinates footnotes and now thinks it can design drugs. The company announced an internal drug discovery program because apparently selling AI tools to pharmaceutical companies wasn't ambitious enough. They want to do the drug discovery themselves. Bold move for a firm whose primary achievement is making a slightly less unhinged version of ChatGPT.
The logic goes like this: train a language model on billions of tokens, watch it predict the next word in a sentence with alarming confidence, then point it at protein folding and hope for penicillin. What could go wrong? Besides everything. Besides the fact that drug discovery takes a decade and a billion dollars and requires actual biochemists who understand why molecules do things instead of predicting they might.
Tech giants already bet on healthcare. Google tried. IBM Watson tried. They all discovered that reading medical journals faster than a human doesn't mean you understand medicine better than a human. But Anthropic watched those companies light money on fire and thought, "Yeah, but our model has Constitutional AI." Constitutional AI will not help you pass Phase 3 clinical trials.
Retail traders will buy this narrative. They'll see "AI" and "drug discovery" in the same headline and imagine a future where Claude cures Alzheimer's by Tuesday. They'll ignore the part where Anthropic has exactly zero FDA-approved compounds, zero clinical trial experience, and zero reason to believe a text prediction engine can replace a century of pharmacology. But the stock won't trade because Anthropic is private, so they'll just buy Nvidia again and call it exposure to the healthcare AI revolution.
The real tell is calling it a "push to sell artificial intelligence tools to drugmakers." They're building the program to prove the tools work. Which means the tools don't work yet. Which means they're hoping Pfizer sees a press release and writes a check before anyone asks for results.
Photo by Brecht Corbeel on Unsplash
Leave a Comment