2024: AI Everywhere—and All the Rest

As 2024 closes, it’s worth pausing to take stock. Reviewing this year’s newsletters, a few things are clear. The AI behemoth continued gathering steam. It also crossed over into new industries and combined with other emerging technologies to quicken their pace. But the year wasn’t only about AI. This year also included cryptocurrencies hitting mainstream finance, Apple dropping its mixed reality headset, robotaxis scaling up, and the launch and recovery of the most powerful rocket since we last visited the moon.

?

Here are highlights from some of the year’s biggest stories.

?

Artificial intelligence. The AI firehose was on full blast again this year. There were real advances in video, reasoning, and interfaces. OpenAI blew minds with Sora, an AI model capable of generating high-definition video clips; controversially evoked the sci-fi classic Her with GPTo’s voice mode; and brought unconfirmed rumors to life with its “reasoning” model o1. But the year was perhaps most notable for the sheer number of people exposed to AI (whether they asked for it or not). Google, Microsoft, and Meta jammed AI into tools and applications with user bases numbering in the billions. Apple is set to follow suit, incorporating generative AI tools throughout iOS. Even relative newcomer, OpenAI, recently reported weekly active users of 300 million.

As the field at the top narrowed—the year was hard on some AI startups—the number of AI models roughly equivalent to OpenAI’s GPT-4 exploded. Now there’s Gemini, Claude, Llama, Mistral, and more. Several of these are open models—most notably Meta’s Llama—though whether we should call them open-source is hotly debated. With the field more crowded, the pressure is on to deliver more products, better products, and perhaps most importantly, the next big leap in capability. Big tech companies upped capital expenditures, and OpenAI and Anthropic raised more funds. Leaders talked about powering data centers with enough electricity to run cities. And Nvidia, the top supplier of AI chips, became one of the world's most valuable companies. Meanwhile, in the midst of all this, investors struck a note of caution: Can AI products actually recoup the hundreds of billions startups and companies are spending?

?

What’s next? Anthropic’s demo of an “AI agent” that can autonomously control your computer to get things done was a clue: You’ll hear more about agents next year. Reasoning models, like o1, are also proliferating, and companies believe they can improve the quality of existing models. But whether scaling up new models and training data will yield more powerful algorithms is an open question (more below). The Information reported in November that OpenAI’s latest scaled up model hasn’t shown the dramatic improvements we’ve come to expect in recent years. While some leaders are unconcerned, others say new scaling laws, like allowing reasoning models more compute to improve their “thinking,” are poised to take over. Whatever happens, it’ll happen fast. Stay tuned.

?

AI+. Arguably, chatbots and agents aren’t the most important story in AI. The approach that has worked so fabulously in language, images, and video appears to be somewhat generalizable—if you have enough data. In a recent example, Niantic said it’s using Pokémon Go data to train what it calls a “large geospatial model” with a deep understanding of the physical world and possible applications in augmented reality and robotics. Indeed, this year saw AI accelerating work on general-purpose robots, and even Boston Dynamics’ humanoid robot, Atlas, long hand-programmed, got in on the AI game. Other areas of note include biotech—where models like DeepMind’s AlphaFold 3 can generate models of biomolecules—and weather prediction—where a DeepMind algorithm recently outperformed classical methods at medium-range forecasting. There will be more. In short, to chart AI’s future, follow the data.

?

Apple reality. Apple makes waves when it enters a new market. This year was no exception when the company released its Vision Pro headset in February. The device is Apple’s best shot at mixed reality, and in many ways, it’s impressive. But with Apple already showing signs it will discontinue the headset, it’s clear people are still unwilling to sacrifice style, comfort, and connection—not to mention pay thousands—for mixed reality devices. All that’s a tall order for today’s tech, but it also points the way forward: More miniaturization. Meta’s prototype Orion AR glasses, demoed this year, offer a glimpse of where the industry, Apple included, is headed. Indeed, Orion’s stripped down predecessor—smart glasses made with Ray-Ban—are already an unlikely hit.

?

Robotaxi rides. Remember when hordes of robotaxis were supposed to hit roads by 2020? That didn’t happen. But companies kept testing, developing, and extending services, and the technology progressed. This year, commercial robotaxis really rolled. From May to October, Waymo tripled the number of weekly rides it operates to 150,000, totaling over a million miles every week. The company also recently opened its services to anyone in Los Angeles. Austin, Atlanta, and Miami are up next. If you’ve ridden in a Waymo, you know they’re for real. Next, we’ll have to see how smoothly they scale.

?

401-Krypto. Bitcoin and Ether officially entered mainstream finance with the approval of their first exchange traded funds (ETFs). This means regular investors can now easily add crypto to their portfolios. Buoyed by the approvals and expectations of easing regulatory scrutiny in the US, Bitcoin broke $100,000 for the first time in December. Crypto as a whole is now worth well over $3 trillion. The industry has had a rough ride in recent years, and much of its value is still driven by swings in investor sentiment. But with new strategies riding on the back of the latest bull, it’ll likely make headlines next year.

?

To the stars. SpaceX launched Starship four times this year. But the big news was when the company landed Starship's booster for the first time in October. SpaceX has been routinely landing Falcon 9 boosters for years, but with twice the thrust of the Saturn V moon rocket, Starship is much bigger and more difficult to land. To accomplish the feat, SpaceX built a pair of robotic arms into the launch tower to catch the rocket after it steered itself home and eased back to the launchpad. There are challenges ahead, including landing the booster after an orbital launch, landing Starship itself, and showing it can be refueled in orbit. SpaceX will begin tackling these next year with a schedule that may include over five times more launches than it totaled in all of 2024.?

More News From the Future

The end of an era? Companies may be struggling to scale new AI models.

Scale struggles. Citing unnamed sources at OpenAI, The Information reported last month that the successor to OpenAI’s flagship AI model, GPT-4, is not showing the big performance gains experienced when they made the leap from GPT-3 to GPT-4. Separately, the tech publication reported Google may be facing similar headwinds with its Gemini model. In both cases, data woes are central to the problem. Previously, AI companies created AI training datasets by hoovering up publicly available information online. More recently, they struck deals for high-quality proprietary datasets—like those behind the paywalls of large publications—added video and image data, and experimented with synthetic data created by existing AI models. If the reports are accurate, these efforts may not be yielding the kinds of results companies had expected.

?

Tapped out? Google and OpenAI built Gemini and GPT-4 with a type of algorithm called a large language model (LLM). When researchers increased the number of connections in LLM neural networks and fed them more training data, they showed impressive performance gains. Known as AI scaling, the strategy worked like clockwork from one generation to the next, leading AI companies to spend billions of dollars to build out enormous data centers stocked with advanced computer chips. If the trend has tapped out for now, future gains are uncertain.?

Nothing to see here. Not everyone believes there’s a problem. At the Cerebral Valley AI Summit, Anthropic CEO Dario Amodei said that, when it comes to training data, “I mostly don’t think there’s any barrier at all.” If he’s right, we might expect performance gains will continue. But even if they don’t, it’s worth noting there’s room for improvement downstream of the core algorithms, even in current models. And there’s a new kid in town: So-called “reasoning” models, like OpenAI’s o1, are showing strength in areas where older models typically struggle. There may even be a new kind of “scaling law” with these models—though it’s costly to customers—where they get better the longer they’re allowed to “think” through a problem. Tack one onto GPT-4’s successor, and the new hybrid model may be more powerful, regardless of gains from scaling.

?

Meta to insure its global traffic with a world-spanning fiber-optic cable.

For $10 billion you could either train a next-generation AI model—or wrap the planet in fiber-optic cable. At least, that’s the price tag Meta’s eyeing, according to TechCrunch, to route 40,000 kilometers of undersea fiber-optic cable—for its own exclusive use—from the US East Coast to South Africa, India, Australia, and back to the West Coast. Possible motives include insuring traffic to global markets remains uninterrupted and avoiding geopolitical hotspots where cables may be damaged or sabotaged. More speculatively, it could allow a high-bandwidth connection to train and run AI using Indian data centers.

?

Patients with lupus go into full remission thanks to CAR-T cell therapy.

A therapy that reprograms the immune system’s warrior T-cells to target and fight cancer—known as CAR-T—has shown very promising results, especially in the treatment of blood cancers. Now, researchers are aiming CAR-T at autoimmune diseases once thought incurable. Progress in one of these, lupus, suggests the strategy works. Over 40 lupus patients have been treated in trials, and nearly all went into remission. CAR-T is expensive and difficult, so researchers are looking for simpler alternatives that get the same results.

?

?

Upcoming Events

Introducing The Future of Biotech

While AI dominates headlines, researchers and startups are leveraging biology to transform manufacturing, supply chains, human health, and material science. These innovations will unlock unprecedented value and disrupt markets. On May 7-9, 2025 in San Diego, we will host experts, innovators, and executives to explore and shape the future of biotech-driven innovation.

Want to learn how biotech will disrupt your industry? Join us in San Diego or virtually.

?

Apply for our 2025 Executive Programs

Starting January 1st, 2025, the investment for our Executive Program will increase from $14,900 to $15,900. However, you have the opportunity to secure 2024 pricing for any 2025 program (April, October, or November) by completing your enrollment this year. To ensure you lock in current pricing:

• Submit your application
• Complete the acceptance process
• Secure your seat with payment by December 31st, 2024?

Ready to apply? Futureproof yourself.

-----------------------------------------

Thanks for reading. We hope you enjoyed this month's updates and found something to inspire you on your exponential journey.

See you next month!

The Singularity Team