Source item
Open original source ↗Model builders overfit for evals that consumers have zero idea how to interpret.
Libraries of important and controversial prompts with side by side comparisons of answers, like the below, are much more interesting and useful.
Quoted context
These are the quoted author’s words.
Grok 4.20 is BASED. The only AI that doesn’t equivocate when asked if America is on stolen land. The others are weak sauce. https://t.co/KEz4MPy2YB
elonmusk ↗