Decoding the Latest Generative AI Products from Google & Meta
With big tech rushing to roll out new products, the AI arms race paradoxically enters its hurry-up-and-wait era
It’s been another busy week in the ongoing arms race of generative AI.
On Wednesday, Google released Gemini, its latest large language model (LLM), and immediately integrated it into Bard. In a unique tri-tiered structure, Gemini encompasses three versions: Nano for Android devices (starting with the high-end Pixel phones), Pro for regular AI tools including Bard, and Ultra for data centers and enterprise applications, which enables a high level of versatility that could theoretically widen its use case and speed up adoption. DeepMind, Google’s AI research unit, also released AlphaCode 2, an advanced code-generating AI powered by Gemini, reportedly demonstrating a remarkable capability in programming contests.
Not to be outdone, on Thursday, Meta unleashed a slew of AI-related updates: celebrity-based AI characters, previously unveiled at its Meta Connect event in September, are now live for U.S. users to chat with on WhatsApp, Messenger, and Instagram; a standalone AI image generator powered by Meta’s Emu model is now available for anyone with a Meta account; and a new “reimagine” feature that allows users in group chats across Meta apps to recreate AI images with text prompts. Moreover, Meta also teamed up with IBM to launch an “AI Alliance”, an industry body to support “open innovation” and “open science” in AI.
All these announcements came merely two weeks after the most riveting power struggle that enveloped the AI world: OpenAI’s star CEO, Sam Altman, was mysteriously fired from the company by the board, and then quickly rehired under pressure from Microsoft within the span of five days. No one is quite sure what led to the firing in the first place, but OpenAI had stopped letting new users sign on for its paid premium tier a week after the major upgrades to ChatGPT it rolled out at its first developer conference, while also postponing the launch of its GPT Store to 2024 without much explanation.
In a way, all these major AI releases coming from Google and Meta this week — in addition to the AI products that Amazon announced at its re:Invent event in Vegas last week — can be read as a reactionary deluge to the small window of opportunity left open by OpenAI’s recent behind-the-scenes drama, allowing the big tech companies to play catch up over the past two weeks.
Yet, despite all the brouhaha surrounding AI releases this week, a closer look reveals that none of them seem guaranteed to challenge OpenAI’s (and Microsoft’s, by association) leading position in the current competitive landscape.
Let’s start with Google. Despite the Alphabet company boasting that Gemini Ultra (its most capable version) beats state-of-the-art results in 30 of 32 benchmarks used in LLM research, outperforms ChatGPT in most tests, and is the first AI model to outperform human experts on a multitasking test called MMLU, which covers 57 subjects including math, physics, law, and ethics, the first wave of user reactions to the new Gemini-powered Bard have not been unanimously great, as some users reported incidents where it gets basic facts wrong, makes glaring mistakes in translation, and, in response to questions about controversial news, simply tells users to Google it.
Skeptics also point to Google’s long, spotted history of starting and then abandoning products as evidence for its potential lack of commitment to the upcoming Gemini-powered products. Moreover, Bloomberg reports that Google’s Gemini video demo of its multimodal capabilities wasn’t carried out in real time or in voice; the model was shown still images and given text prompt, furthering sowing doubts of the model’s true capabilities.
Of course, it’s worth keeping in mind that Bard is currently powered by Gemini Pro, which isn’t the most capable version of Gemini. Google says Gemini Ultra is set to launch sometime next year in Bard and other Google products, which would likely then boost the accuracy of Bard’s answers. Still, one could make the argument that Google is making a strategic mistake by integrating the less capable versions of Gemini into Bard and risking users forming a negative first impression of the new Gemini model. After all, a layman consumer is far less likely to care for the difference between Gemini Ultra and Gemini Pro. In fact, reports from Monday suggested that Google was supposedly postponing the Gemini launch to early next year, so the Wednesday launch came out of the blue for many. Now, judging by the early user reactions, perhaps it could’ve been more impressive to launch with Gemini Ultra in January when it’s ready.
Similarly, Meta’s rollout of its previously announced AI features this week also felt too minor to really move the needle. By focusing on image generation tools in chats and AI character chatbots, Meta’s approach still squarely lands in the “generative AI as an entertaining toy” territory rather than positioning it as a powerful creation and productivity tool. Granted, it is an approach that makes perfect sense for Meta, considering the company’s two top concerns are, one, to keep users engaged and entertained across its apps, and two, to avoid opening the floodgate of misinformation that AI-generated content could bring, or at least not be held responsible by regulators for it (see: the aforementioned AI Alliance with IBM). Still, bound by these two concerns, it would seem unlikely that Meta alone would move the needle on generative AI development by much.
That said, it wouldn’t be fair to completely count out Google and Meta from the AI arms race yet. After all, both companies own a host of enormously popular services and products, with the likes of YouTube, Instagram, and WhatsApp, each with over 2 billion users worldwide, which could easily be leveraged to scale AI-powered features and spur mainstream adoption. At a time when AI-generated images are sweeping through social media and fandoms — for example, the “make it more” trend or the Gag City memes — there is certainly value for Meta to launch a standalone AI image generator that anyone can use for free, and for Google to enhance Bard’s multimodal capabilities. Still, in order to win over users from competitors, the AI models deployed by both companies will have to get a little better. Based on my own experience trying out both Gemini Pro-powered Bard and the Emu-powered Meta image generator today, it would seem that both still have some catching up to do.
So, to recap, OpenAi is still leading the generative AI market, but Google might be catching up soon when Gemini Ultra rolls out in January, and Meta’s quasi-open-source approach will see a real test in the coming months in terms of keeping up with the competitors. As with many hot innovation territories, there is a lot of hurry-up-and-wait happening while figuring out the right use cases and the best product-market fit. And, like many things in life during this time of the year, it is time to circle back after the holidays.
