TLDRocket
Sign in

LLM-powered Biographies

Eugene Yan

A data scientist asked six different AI models to write his own biography, just to see what happened. The results ranged from mostly-right to wildly fictional, showing how much these models just guess.

Based on reporting by Eugene Yan — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

Eugene Yan decided to run a little experiment on himself. Curious about what large language models actually know versus what they confidently invent, he asked several of them, from OpenAI's GPT-4 down to Cohere's command-xlarge, to write his biography. He's a real person with a real, googleable career, so this doubled as a decent litmus test for how these systems handle facts about someone who isn't Einstein or Elon Musk.

GPT-4 came closest to reality, nailing his first two jobs but inventing degrees from the London School of Economics and NYU Stern that don't exist, plus a fictional stint at Amazon. GPT-3.5-turbo did worse, giving him a Stanford master's, a Google job he never had, and a startup called Axon he never founded. Text-davinci-003 abandoned reality entirely, turning him into a Toronto-born venture capitalist who backed Uber and Dropbox. Older davinci from 2021 didn't even attempt a biography, instead spitting out a broken, repetitive list of

My take — AI-written commentary, not fact-checked reporting

favorite loves

Read more about this at: Eugene Yan

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.