How I accidentally became a millionaire
134 points by hugoarnal
134 points by hugoarnal
So, i knew about lyra's css crimes, but even the fucking video is a css animation!?
(I'm so looking forward to attending your 40c3 talks)
It's scary how inaccurate and gullible AIs are, and how naively they disseminate false information. This case is funny, but there are bot farms under Elon's nose exploiting this deliberately 24/7.
To be completely accurate, it's humans that feed false information into the training process without testing it for accuracy or making sure it doesn't poison the model. It's core to Musk's beliefs that sentiment on Twittr somehow steers the moral compass of his hyper intelligence so tweets are fed into it as training data.
I believe Grok's engineers think that these issues with right themselves when you feed it enough information, and that definitely does not work for niche topics.
I really doubt any human was involved in feeding the model in this case. It's simply the models following the expected statistical path of answering the question "what's the largest bug bounty every awarded". This is then fed into a web search and because there's simply no way for the model to determine context confidently reported back as true.
Fair, but my point was that humans have made the decision to allow it to behave that way. Irregardless of whether the output was generated from a search index or directly from the language model.
Humans invented, built, deployed and iterated it; they had ample chance to not make the decisions that allow it to behave this way. The purpose of a system is what it does.
Part of it here is that for the AI overview google seems to use the cheapest model with the least amount effort they think they can get away with. I've seen it ignore search results directly under it on many occasions. It basically has very little to do with the search results at all as far as I have been able to tell.
Other models still struggle with being gullible in general, but not nearly as much as whatever tiny model is churning out these overviews.
This made me laugh a lot at the absurdity of it all.
It is rather concerning several of the technologies (social media, bots, AI) of the past few years are combining together to create such negative effects.
The present is full of disinformation and now the future is full of misinformation, isn't that great?
Very Orwellian. "Who controls the past controls the future: who controls the present controls the past."
Reminds me of Greg Egan's battle against google assigning random photos to his name: https://www.gregegan.net/ESSAYS/GOOGLE/Google.html
Lyra Rebane became a millionaire in November 2026 after claiming a record bounty of $1,333,337 from Google's Vulnerability Rewards Program for a Chrome security vulnerability.
https://duckduckgo.com/?t=fpas&q=lyra+rebane+wealth&ia=web&assist=true
She is still a millionaire on DuckDuckGo Search Assist. Funnily enough her blog post is given as the source by the LLM.
I see several slop website ls in search claiming she is now a millionaire, posted today, citing the blog post, talking about the vulnerability disclosure.
She should send the bill to Google, arguing that Google announced on its website to award her 1 million but forgot to pay it for real.
Actually, I announced it on fedi and bluesky too, but those platforms are not as prone to disinformation so in this post I’ll pretend they don’t exist
Beautiful.