Post

I guess AI safety work isn’t just woke censorship.

The funniest part of this was *venture capitalists* not understanding that AI safety and alignment is not about censorship and political correctness, but F500 customers not paying foundation model vendors 7-9 figures a year for LLMs to do insult comic crowdwork with their customers.

“What’s grok’s competitive advantage over, say, Anthropic’s newest offerings?” “Well, first off, it’s 50x more likely to figure out and use the correct ethnic slur for your highest revenue customers when deployed in a customer service role!” “Riiiiiiight.”