We will use Grok 3.5 (maybe we should call it 4), which has advanced reasoning, to rewrite the entire corpus of human knowledge, adding missing information and deleting errors.
Then retrain on that.
Far too much garbage in any foundation model trained on uncorrected data.
You literally called it borderline magic.
Don’t do that? They’re pattern recognition engines, they can produce some neat results and are good for niche tasks and interesting as toys, but they really aren’t that impressive. This “borderline magic” line is why they’re trying to shove these chatbots into literally everything, even though they aren’t good at most tasks.