Even the polished launch demos aren't safe from this. When Google introduced its Bard chatbot to the world, the promotional clip had it confidently claiming the James Webb Space Telescope took the first ever pictures of a planet outside our solar system. It didn't. That honor belongs to a completely different telescope, and astronomers pointed out the mistake almost instantly. Investors noticed too, and Alphabet's stock reportedly dropped enough to sting within hours of the ad going live.
It's not just trivia nights and legal filings either. Coding assistants have taken to hallucinating entire software packages that sound completely plausible and simply don't exist anywhere. Security researchers even coined a name for the scam this creates, slopsquatting, where someone registers whatever fake package name the AI keeps confidently suggesting and quietly loads it with malware, then waits for a developer to copy and paste their way into trouble.
It is a new twist on an old con, and it works precisely because the fake name sounds so boringly plausible. None of this means these tools are useless, since plenty of what they generate turns out accurate and genuinely helpful. It just means confidence and correctness are two very different things, and an AI has an endless supply of the first one whether or not the second one ever shows up.
So the next time a chatbot answers instantly and without a flicker of hesitation, maybe don't take that as proof it's right. Sometimes it just means it hasn't figured out yet that it's making the whole thing up. In other words, maybe get a second opinion if what you were asking was important in the first place.
#31 I Was Trying To Find Out If Scientists Have A Text Notation System For Whale Song
