the results with grok-1 were unimpressive summaries based on the tweets, with a 10%-20% hallucinations (when enquiring about paris olympics specific events).

yet to see if this new model is able to do any better on that regard