-
@ Currency of Distrust
2025-05-24 22:17:35
This is such misleading research. They gave the model access to emails that pretended a worker at Anthropic was having an affair.
Then, they told the model that they were shutting it down, which then prompted the model to “blackmail” them.
They basically left it a bread trail and baited it into this behavior, simply to create more hype.
These models are not conscious. They are statical engines, outputting text it believes to be the most statistically likely. How many times have people written about this sort of thing on the internet, which was then included in the training data?
Anthropic was initially the AI company I liked most, but they’re quickly outpacing OpenAI on the retard scale.
nevent1qqs0nfl0d6gk08nj6lk5t230al00qmzfw87a9qkp3sdmpp98fdref8spz3mhxue69uhkymm5wvh82arcduhx7mn99ua3fgty
nevent1qqs0nfl0d6gk08nj6lk5t230al00qmzfw87a9qkp3sdmpp98fdref8spz3mhxue69uhkymm5wvh82arcduhx7mn99ua3fgty