A tool asks 14 LLMs who you are. Most invent someone who never existed
George McBay typed his name in and learned he was an African-American chemist, then a Big Lebowski actor. Hundreds of HN users did the same and met strangers wearing their names: rugby players, French diplomats, white nationalists, an Al-Qaeda bomber.
Show HN "Are You in the Weights?" fires your name at a dozen-plus models, from Llama 3.2 1B to GPT-5.5, and grades how 'memorized' you are with a strength score and a 'Top N%' rank. The replies read like funhouse mirrors: people with a genuinely unique name still got fake biographies, fake sports careers, fake crimes.
Then a privacy warning surfaced: every name typed in lands on a public 'latest' leaderboard that paginates endlessly. And one quiet question nobody answered: what does being in the weights mean for the right to be forgotten?
A name goes in, a stranger comes out
The Show HN "Are You in the Weights?" does one thing: you type a name, and it queries a stack of large language models, from tiny Llama 3.2 1B up to GPT-5.5, Gemini, DeepSeek, Kimi and more, asking each who that person is. It then scores how strongly the name is 'baked into' the weights, a yellow number it calls strength, plus a Top N% rank, and flags the answers it judges to be hallucinations. The pixel-art portraits and retro-game styling were a thread favorite. The substance underneath was less flattering.
The biographies are mostly invented, even for unique names
The pattern that emerged across nearly 200 comments is that the tool is a hallucination generator dressed as a memory test. People with a one-of-a-kind name, who should be easy to get right or admit ignorance about, instead got confident fiction.
"I am, as far as I know the only person in the world with my name… this tool tells me I am, for example, a right wing politician, a journalist, and a researcher on solar fuels." — an engineer who is the only person with his name
"Llama 3.2 1B says American actor, best known for his roles in films such as 'The Big Lebowski' and 'The Big Lebowski 2'. Nailed it! /s" — another user who ran it on himself
The darker collisions piled up: a French diplomat who was ambassador to Congo, a "prominent white nationalist activist," an Al-Qaeda member tied to the 2004 Madrid bombings. A 1-billion-parameter model even rewrote Mozart's biography with the wrong dates. One reply summed up the failure mode precisely.
"Even the leading models still very much hallucinate esp when they don't have the ability to go looking for more context. It took various things related to stuff I work on but mixed them up and added pure invention…" — a developer with 22k+ karma who saw his own work mangled
Who the models actually remember
The hits clustered in one place: people whose online handle and output sit in the open-source and writing corpora the models trained on. jedberg, the former Reddit and Netflix engineer, scored top 5% and saw most models get him right (one still insisted he was a VP at Datadog). A long-time open-source contributor noted the bubble effect: "being in the Open Source world is a bit of a bubble as far as the weights are concerned." Models like Kimi and DeepSeek picked up Medium and blog writing more than the others. The takeaway is that the score measures corpus presence, not fame, and certainly not accuracy.
The leaderboard you didn't know you joined
The sharpest reversal wasn't about accuracy. It was about what the site does with the names.
"PRIVACY WARNING: Every name/text entered into this site is publicly listed on the 'latest' leaderboard which seems to paginate endlessly." — the commenter who found the public leaderboard
Others piled on: no privacy policy, a surprising number of tracking cookies for a toy, and a blunt verdict from one reader: "Only a fool would enter their name in this." A couple of people read it as a clever funnel to harvest real names against IP addresses.
The unanswered question under the joke
Most people came for the ego check, scored themselves, and laughed at the absurd misses. But one comment, left almost in passing, is the one worth keeping: if your Reddit history and old blog posts are demonstrably in the training data, immortalized without consent, then the model's confident wrongness about you is permanent too. As Evan Tahler, who works on AI tooling at Arcade.dev, put it after confirming he was in the weights: "So… what does this mean for the right to be forgotten?" The tool answers a fun question and accidentally raises a serious one: you can be both memorized and misremembered, with no way to correct either.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
An essay says 'start AI over from zero.' The thread flipped it apart
The post argues consent vanished the moment AI entered the room, and demands the whole industry restart with consent as a core concept. Within an hour HN had reframed the whole thing: the anger isn't about machine learning at all.