The hotdog test that makes big tech trip over itself

A BBC journalist writes on his blog that he’s the world champion hot dog eater. No fancy hack, no deepfake, just plain old bullshitting on the internet. Within a day, ChatGPT and Google’s supposedly “intelligent” systems parrot his nonsense like it’s gospel. One system, Claude from Anthropic, smells the bullshit. And what does this teach us? That all these companies launched their systems while knowing damn well they’re as manipulable as a ballot box in a banana republic.

The solutions have been gathering dust for years. Multi-model verification where systems check each other. Source evaluation that understands one obscure blog isn’t proof. Uncertainty quantification that says “hold on, this doesn’t add up.” But those layers cost computing power and time, and in the race for market dominance where every month’s delay costs billions, accuracy was sacrificed on the altar of speed. Because fuck the truth, right? As long as the stock price keeps climbing.

OpenAI talks about “safety” while racing to pump ChatGPT out to billions of users. Google dropped “Don’t be evil” because it got too difficult to get rich and stay honest. Anthropic plays the modest underdog while fighting just as hard for the same venture capital millions. This isn’t a competition for the best product, this is a land rush where whoever plants the flag first rakes in the profits.

You use these systems every day. For medical advice, financial tips, legal questions. And you know they lie, hallucinate, sell crap as facts. But it’s so damn easy, so temptingly fast. We have the technology to fix this, but not one company implements those solutions before they ship. Why would they? The users keep coming anyway.