3 ms·
Perhaps the smaller model used in o1 is over trained on arxiv and code relative to 4o (or undertrained on legal text)
by m101 2y ago
Perhaps the smaller model used in o1 is over trained on arxiv and code relative to 4o (or undertrained on legal text)