AI Heat Wave

If you thought the AI heat wave was cooling, think again. This month brought a flurry of developments. Google released its most powerful AI model to the public—Gemini Ultra—only to upstage itself a week later with Gemini Pro 1.5. The latter matches Ultra’s performance and includes a huge context window—the amount of data you can enter in a prompt—for video, audio, and text, while using less computing power. Google is weaving Gemini into products that reach billions and launching subscriptions.

Not to be outdone, OpenAI previewed Sora, the most advanced text-to-video AI model yet, outshining Lumiere, a similar model Google announced just weeks earlier. Sora generates stunning high-definition videos from simple prompts, and in a technical report, researchers made the case that, scaled further, Sora’s ability to model 3D spaces with changing camera angles as well as cause and effect could make it a useful “world simulator.” It’s amazing progress. Not quite a year ago AI video models went viral for generating trippy beer commercials and weird videos of Will Smith eating spaghetti. 

Driven by its OpenAI partnership and AI-first strategy, Microsoft has now overtaken Apple as the world’s most valuable company with a market capitalization of $3 trillion, and Nvidia, the supplier of the most popular AI computer chips, has leapfrogged Alphabet. Meanwhile, OpenAI recently closed a deal valuing it at $86 billion, nearly triple its valuation 10 months ago.

But recent announcements are as much about the business of AI—competition to impress or reassure investors, drive media coverage, and sell products—as they are about advances in the core technology. Google is still playing catch-up with OpenAI and Microsoft, who took the lead with ChatGPT, GPT-4, Copilot, and subscriptions last year.

The big question now is whether enough people will actually use and pay for these new tools. The jury is decidedly out, but reviews from enterprise early adopters of Microsoft’s Copilot have been mixed. With Gemini, it’s Google’s turn. Both companies can afford to experiment, but eventually, they’ll have to show there’s a sustainable business model. 

As for where the technology goes next, reading between the lines has become increasingly challenging as the industry gets more competitive and less transparent. But there are a few clues pointing to two trends that will likely matter this year and beyond.

Scale. Building ever-larger models that consume more data was the industry’s north star leading into the current boom, and according to AI leaders, it still matters. OpenAI took care to emphasize that further scaling video models like Sora would yield impressive gains. Elsewhere, Amazon said scaling text-to-speech models likewise bears fruit, making computer-generated voices far less robotic. But perhaps the most obvious (and eye-popping) signal AI needs more computing power was a report that OpenAI CEO Sam Altman is hoping to raise a lot of cash to make AI chips. The ballpark amount could be as high as $7 trillion—all in, from many players—over a number of years.

Design. But computing power isn’t everything. In an interview with Wired’s Will Knight, Google DeepMind CEO Demis Hassabis suggested that to realize future step changes, the industry will need to go after “maximum scale” and come up with more innovations in the design of AI software. He said adding reinforcement learning, DeepMind’s “bread and butter,” to existing models could yield new capabilities like planning and reasoning. Other innovations may include the use of synthetic data or the addition of rule-based AI.

What, then, does the next big step look like? To Hassabis, it’s simple. Today’s most advanced AI algorithms are basically “passive Q&A systems,” he said. Think of them a little like playing Ping-Pong against a wall. You have to keep serving the ball for the game to continue. The next step will be AI agents that, once given a task, can keep the rally going by themselves until its complete.

“We're investing a lot of time and energy into that area, and we think that it will be a step change in capabilities of these types of systems—when they start becoming more agent-like,” said Hassabis. “We’re investing heavily in that direction, and I imagine others are as well.”

Indeed, rudimentary agents already exist, and OpenAI is reportedly working on agents that would operate your computer or browser. “Eventually, you’ll just ask the computer for what you need, and it’ll do all of these tasks for you, “ Altman said in November.

More News From the Future

The Apple Vision Pro is here. Is it here to stay? 

Space launch. After announcing the Vision Pro mixed reality headset last June—or as the company prefers, “spatial computer”—Apple began delivering devices this month. The Vision Pro is the first new hardware product from the iPhone maker since the Apple Watch. Though expectations for sales of the pricey headset weren’t high, reports suggested an initial pre-order run of 200,000 sold out. Fans were spotted wearing the device out in the wild ordering lattes and walking down the street, as hot takes and prognostications flooded the internet.

Consensus? Professional reviewers varied in their conclusions, but most agreed on a few points. The Vision Pro’s display, design, and ease of use are impressive; it’s also heavy and uncomfortable to wear for long periods of time. Though few expected the device to be a mainstream hit due to price, its debut is still being watched closely. Apple is rarely first to the punch, but its design choices can define categories. Virtual reality headsets, available for years, are still niche. Can the Vision Pro and future iterations open the category to more consumers?

Work in progress. It’s still unclear what would induce people (beyond VR fans) to buy a Vision Pro. People have suggested killer apps like work, movies, mindfulness, and kitchen timers. Obvious points in need of fixing include price, which can balloon to $4,600 with add-ons, and limited apps. Some solutions are perhaps already on the way. The app store has reportedly tripled in size. Others will be tougher to nail down. If the ultimate goal is a low-profile, comfortable pair of glasses that can match the capabilities previewed by the Vision Pro, then Apple’s cutting-edge device suggests mixed reality for the masses is still distant.

Open-source AI is hot on the heels of private industry leaders.

Open-source models continue to advance quickly, with some algorithms closing in on GPT-4 levels of performance. Meta released an open-source coding sidekick said to rival private AI models, and Hugging Face launched an open-source “AI app store” alternative to OpenAI’s custom GPT store. Some have noted open-source models, like Meta’s Llama 2, aren’t as open as they could be—AI2 remedied the situation with its OLMo models, which are far more open by comparison and licensed for any use including commercial.

SpaceX will use a single Starship to launch a space station.

SpaceX is still developing its next-generation Starship rocket, but that hasn’t stopped the company from signing up future customers. One of these, a joint venture between Voyager Space and Airbus, is building the Starlab space station. The fully assembled space station, which will be half the volume of the International Space Station, fits inside Starship’s payload bay. By comparison, it took 42 flights to assemble the ISS in orbit. It’s a great example of the radical leap in launch capacity a completed Starship represents. 

Lab-grown meat is still costly to produce. Researchers are on the case.

Cultured meat has long promised to reduce the environmental footprint of livestock by doing away with live animals and growing meat from the cells up. But despite years of development, the idea has yet to take off, due partly to cost. To slash expenses, a recent study genetically modified cow muscle cells to produce their own growth factors, which make up 90 percent of production expenses. In another study, scientists grew beef cells in grains of rice. They estimated the beefy rice could cost as little as a dollar per pound.

Upcoming Events

EXECUTIVE PROGRAM

Join us for a LIVE Executive Program information session

In this 30-minute webinar hosted by Singularity VP (and AI expert) Alix Rübsaam, you'll get to know more about the program, ask questions, and meet a recent Executive Program graduate.

Register for March 28 at 11am ET

Ready to apply? Choose one of our 2024 EP dates:

April 7–11, 2024

October 6–10, 2024

November 10–14, 2024

ABUNDANCE360 

Applications for The Abundance360 Summit are closing soon!

Hosted by Singularity co-founder Peter Diamandis, The Abundance360 Summit will be held March 18–21, 2024. Learn more & apply.

THE FUTURE OF AI 

Early bird registration is now open

From May 8-10, we're convening the sharpest minds in AI at the Computer History Museum in Mountain View, California. Learn more and secure your early bird seat before March 8.

Thank you for stepping into the future with us! Join our global community of over 200,000 futuremakers - sign-up and unlock early access to the Singularity Monthly newsletter and discover the most impactful technology breakthroughs - get a glimpse of tomorrow, today.

The Singularity Team