Microsoft is actively using artificial intelligence with its updated Bing search engine and Edge web browser, both of which are based on what appears to be exclusive access to the successor to the popular OpenAI ChatGPT language model.
The new AI is as yet unnamed, described only in a blog post as follows:
…OpenAI’s next-generation major new language model, more powerful than ChatGPT and configured specifically for search. It builds on the core knowledge and achievements of ChatGPT and GPT-3.5, and it’s even faster, more accurate, and more efficient.
“Configured specifically for search” was echoed by several executives at a Microsoft event, so it must be some coherent language that somehow doesn’t overestimate the capabilities of the model. They called it a “big new language model,” though it’s hard to say how much of a step up from previous OpenAI models. There has been speculation that it was GPT-4, but so far that term has not been used. I have asked OpenAI for more information and will update it if I get a response.
One of the disadvantages of large models like this is the enormous processing power required to run them, causing many potential ChatGPT users to wait several minutes before starting a session. Microsoft’s focus on scaling makes sense, especially since it will certainly foot the bill for computing resources and servers.
Unlike ChatGPT and other GPT models, the AI-powered Bing is accessed directly through a normal search interface, and the security system Microsoft created is called Prometheus. Of course, this is an ominous name, because although Prometheus was the god of wisdom and cunning, he also famously ended up in endless torture, chained to a rock with a huge bird forever pecking at his guts. Let’s hope that Bing doesn’t end up the same way.
Prometheus is part of a layer of security and control around the model that acts as a disinfectant and filter, keeping track of obviously inappropriate or incorrect results. But it also inputs relevant data, such as location, context and relevant information, to adjust or improve the inputs and outputs of the underlying model.
The next-generation model has also been applied to Bing’s search rankings, “resulting in the biggest jump in relevance in two decades.”
