2 ms·
There is nothing new here in principle. The difference is that before, you had to convince GPT-4, with cleverly crafted prompts and/or parameter tricks, to ask
by TeMPOraL 3y ago
There is nothing new here in principle. The difference is that before, you had to convince GPT-4, with cleverly crafted prompts and/or parameter tricks, to ask you to do something for it. What we have now is in principle the same, except OpenAI picked a specific format for GPT-4 (and GPT-3.5) to make requests to you, and fine-tuned the hell out of the model so it's much more eager to use it, and much better at incorporating it into its "reasoning".