Astra fails LaughBench, scoring a 0/20. LaughBench tests whether or not AI models can reliably write jokes about current events that make me laugh, on the assumption that killing everyone on Earth will be somewhat more difficult than making me chuckle. No model has broken the zero laughs threshold yet.

The benchmark is basically a joke, but even though it's a joke I still take it more seriously than the majority of benchmarks for measuring something like IQ, so I'm keeping an eye on it.