Microsoft, ChatGPT, and OpenAI: One month later

Almost exactly one month ago, I wrote about how the success of ChatGPT and Microsoft’s (almost concerningly) close relationship with OpenAI is going to put pressure on rival firms to release their language models and might mark a turning point in generative AI.

A few things of note have happened since then:

• In an unexpected turn of events, sensationalist media have sensationalised the potential of GPT-3 and its planned successor–to the point where OpenAI CEO Sam Altman has himself stated said that people are going to be disappointed by GPT-4.

• Microsoft’s gotten even closer to OpenAI, recently announcing a new ‘multibillion-dollar investment’ in the firm. Much of that money will go back to Microsoft in the form of Azure bills, but it’s still clear that Microsoft sees tremendous value in generative AI in both enterprise and consumer software.

  • They’ve launched ‘Teams Premium’, which includes, amongst other features, automatic generation of notes, tasks, and meeting highlights using GPT-3.5.

  • Screenshots of what will most likely be Microsoft’s ChatGPT-powered version of Bing that leaked yesterday, which include branding of Bing as an ‘AI-powered search engine’, and which show a version of ChatGPT that–as was speculated on this blog and by several others–pulls data directly from the web to generate responses, and, more importantly, is capable of citing its sources.

In what’s been a very successful January for Microsoft–their stock is up 12.77% over the past month–there’s one rival of theirs that is conspicuously absent from the mainstream conversation about generative AI.

A rival which received its first pull request from its cofounder in years; a rival which seems increasingly pressured to disregard the ‘reputational risk’ of generative AI that had previously concerned them; and a rival that’s already under the scanner by the US government over antitrust concerns.

These haven’t exactly been the best few weeks in Google’s history.

This is perhaps the closest they’ve ever gotten to being existentially threatened, as their core products–search and advertising–are under fire from both regulatory bodies and previously dormant competitors, even as they experience a post-pandemic lull in revenue.

That’s not to say that this is the end of Google, or that Microsoft will completely wipe out their search product. At this point, it may be hard to remember that Google were the ones to come up with the ’transformer’ machine learning architecture that enables LLMs in the first place (the ‘T’ in GPT), and that they’ve been showing off their generative models–albeit in highly controlled environments–at their annual IO event for years, long before LLMs had entered mainstream conversations.

And they’ve started working on a strategy to maintain their dominance. They just invested $300M in an AI research lab founded by ex-OpenAI employees, and their CEO has stated their intent to “rethink and reimagine” search, making their language models “available, starting with, LaMDA so that people can engage directly with them”.

These are bold moves for a company that’s until very recently shied away from entering their AI into the public sphere–whose AI research has already been rocked by several internal scandals–and are moves which they almost certainly would not have made were it not for the tremendous pressure being created by Microsoft and OpenAI to integrate conversational AI into their products.

If there’s one consolation for Google in everything that’s hit them this month, it’s this: by the time IO 2023 rolls around, they’d probably be a lot less worried about fighting legal action against their monopoly on search. Because that monopoly might just be over.