Given that they've shown that they don't shy away from making questionable changes to Grok's main system prompt (i.e. the one telling it to not worry about being politically incorrect which subsequently lead to it being able to be pushed into making extremely antisemitic comments)[0], I would hope any company approaches this with that wariness in mind.
The nonprofit Arc Prize says that Grok achieves a new state-of-the-art score on its ARC-AGI-2 test — another difficult benchmark that consists of puzzle-like problems where an AI has to identify visual patterns — scoring 16.2%. That’s nearly twice the score of the next best commercial AI model, Claude Opus 4.
simianwords · 1d ago
How can I try out grok 4 heavy? I can't find API or openrouter access.
ksynwa · 1d ago
What's the model that the @grok twitter account uses?
bravetraveler · 1d ago
:latest
I kid but it's roughly that predictable; whatever was pushed last.
[0]: https://techcrunch.com/2025/07/09/x-takes-grok-offline-chang...
No comments yet
I kid but it's roughly that predictable; whatever was pushed last.