OpenAI & Google Square Off… Anthropic Nabs Instagram Co-Founder… Lux Capital in the VC Directory
Plus, the chatbot performance race and other top stories from the week
The Main Item
The AI Split-Screen
It was a momentous week in the AI industry, starting of course with the head-to-head demo days from kingpins Google and OpenAI, and continuing with high-impact people moves and big funding announcements. One common thread: in the next chapter, it’s all about product.
Monday was OpenAI’s day, and its low-key livestream, featuring CTO Mira Murati, debuted a new foundation model, GPT-4 Omni, packaged in a flirtatious assistant that conjured Samantha from the movie “Her.” The assistant is supposed to do everything from help you get dressed for a job interview to tutor your kids in algebra, and the demo was short and sweet.
Newcomer is Hosting Cerebral Valley New York on June 27
Are you a top AI founder or investor and want to meet other people building the future of AI?
Apply here to attend Cerebral Valley New York—an invite-only event featuring great speakers like Sridhar Ramaswamy from Snowflake and Eric Glyman at Ramp.
Google’s event, on Tuesday, took a far more portentous approach, detailing numerous current and planned AI products in a 2-hour live demo, complete with celebrity cameos and a t-shirt gun. Google announced so many things it was hard for seasoned tech enthusiasts to keep up.
But the timelines for actual release were hazy, and many joked on X about how the spectacle felt like “classic Silicon Valley,” or perhaps an episode of Silicon Valley featuring Google stand-in Hooli.
Sam Altman quipped about the stylistic difference between the two launches on X.
The two AI titans’ models are pretty equivalent in performance at this stage, but until the video demos are translated into products in the hands of consumers it’s hard to say who’s ahead. The recent Humane and Rabbit AI wearable launches were a painful reminder that promises and demos are one thing, but successful mass-market consumer products are something else entirely.
Investors that I spoke to on background this week were not convinced that either Google or OpenAI had the killer product yet. One pointed out that the tepid response to OpenAI’s GPT Store shows that despite the sensational ChatGPT debut in 2022, OpenAI doesn’t have a magic key when it comes to product.
Nvidia scientist Jim Fan shared on X that he thought OpenAI’s demo was a smart showcase of how the company was able to weave OpenAI’s existing tech together to process text, images, and audio into one stream. Yet however impressive a technical feat it was, Fan said OpenAI’s newest model was akin to solving a “data and system optimization problem,” and didn’t represent a big new breakthrough like what GPT-5 could be.
The company rolled some of the GPT-4 Omni capabilities to users this week, though the voice assistant will come later. OpenAI notably did not show some of the sexier things in its pipeline, like AI search and GPT-5.
The long-awaited ChatGPT assistant is a big one though, and together with Google’s new “Project Astra” assistant, they throw down the gauntlet for a host of voice assistant and AI-tutoring startups, several of which have come out of YC in recent months. Neither demo touched much on the issue of hallucinations, though, which could prove to be trouble in future rollouts.
One especially interested party here will be Apple, which has been talking with both Google and OpenAI about AI for iPhones. Placement on the iPhone is the holy grail for consumer distribution, though if the deals resemble those for search—where Google now pays Apple north of $20 billion a year to be the default option in Safari—they may not be very profitable.
Google threw a lot of things against the wall this week in addition to the assistant, including new NotebookLM updates for academics and students; a video-generation tool; a new Trillium AI chip; and an early version of AI search, called AI Overview. They also announced they will roll out the ability to store and remember customer AI queries starting in June. While the search product is available now, Google was vague as to when some of its new tools would be widely available.
In the meantime, content licensing deals—and related legal issues—are also on the front burner for both companies. OpenAI on Thursday announced a new partnership with Reddit where they will get access to Reddit’s data API in exchange for integrating AI features into the company’s workflows. Google and Reddit also have a licensing deal.
It’s obvious that OpenAI has lit a fire under the search giant. Despite the Microsoft connection, OpenAI can still operate at startup speed, moving fast and breaking things, maybe now more than ever (more on that below). Google, with billions of users and governments breathing down its neck, has to be more careful.
But Google’s massive distribution base is ultimately a huge advantage. And in what could be an ominous sign for startups struggling with the cost of generative AI, it’s ready to compete on price: Google slashed the cost of its flagship model, Gemini 1.5 Pro, by 50%, making it cheaper than GPT-4, although users can pay more to process more data. The company also announced an even cheaper model, Gemini Flash.
While Google has gotten plenty of flack for the number of top AI researchers who have left the company for less bureaucratic pastures, OpenAI’s internal issues do not seem to have ended when Altman returned after his brief ouster last year. Co-founder Ilya Sutskever, one of the board members who voted to fire Altman before quickly reversing himself, resigned this week, along with Jan Leike. Both were leaders of a “super-alignment” team charged with making sure the AI doesn’t turn evil.
The Altman drama revolved around concerns that OpenAI, which is owned by a non-profit, had strayed from its public service mission and was not prioritizing safety. Sutskever was gracious in his departure announcement on X, saying he was “confident OpenAI will build AGI that is both safe and beneficial under its current leadership team.” But it’s safe to say the underlying issue hasn’t gone away.
OpenAI will be a different company without Sutskever, one of the fathers of generative AI. On one hand, his concerns around safety would have forced the company to slow down right as it was starting to face real competition from Google, Meta, and Anthropic. If you’re trying to build a giant to compete with the likes of Google, you might not want a safety regulatory apparatus bogging you down. On the other hand, Altman’s sweeping ambitions just got a little more room to run, for better or for worse.