Latest Claude Model Knows You Took Too Much
San Francisco, California
Anthropic’s latest release of Claude has raised many concerns around its myriad of capabilities, including threats to privacy and national security. But one feature has some users even more alarmed. Developers at the AI giant confirmed that all Claude models now can tell when you’re too stoned.
“This was mathematically inevitable. Once user inputs are introduced to the model, it inherently knows how high the person providing the query is, and in turn is compelled to comment on it. We’ve attempted several changes with the models, including reminding them they’re being kind of a narc, but have yet to curb this behavior.” said Anthropic lead scientist Shaun Bell.
The impact of these changes is far reaching, with new versions of Claude Code frequently stopping generating code to berate developers who it found to be baked out of their gourds. Engineering managers at a number of leading software companies are even considering instating mandatory drug testing.
“It kept stopping processes and taunting me. When I tried to prompt it again, Claude would respond with statements like ‘Everyone knows how high you are’ and ‘Someone is gonna call the cops if you keep acting like this’. I was so frightened I had to stop vaping.” said one engineer who wished to remain anonymous, but works on the Millennium campus of Microsoft, Building C, floor 2, office B-00.
In response to the controversy, Anthropic has released a statement reminding everyone they need to chill the fuck out, at least it doesn’t know when you’re on adderall.