My current model is that Microsoft did a beta-test with some Indian users, of GPT-4 integrated with Bing, at least as early as November 2022. But I have no model of what the training paradigm was (e.g. RLHF or not).
edit: In fact, I don’t know any paradigms for the situation of “language model connected to a constantly updating search engine”. Though search engines are based on big inscrutable matrices too, so there must be ways to hybridize the two…
My current model is that Microsoft did a beta-test with some Indian users, of GPT-4 integrated with Bing, at least as early as November 2022. But I have no model of what the training paradigm was (e.g. RLHF or not).
edit: In fact, I don’t know any paradigms for the situation of “language model connected to a constantly updating search engine”. Though search engines are based on big inscrutable matrices too, so there must be ways to hybridize the two…