April 20, 2024

Dream on, Mark Zuckerberg. Your new AI bet is a very long shot

Meta CEO Mark Zuckerberg, who runs one of the largest AI research efforts out there, wants to run one that’s even bigger. It’s a pretty crazy idea.

On Thursday, he said Meta is stepping up its work to address not only artificial intelligence, but also what is known as artificial general intelligence. AI and AGI are already very confusing terms, but simply put, with “general intelligence” systems, Zuckerberg wants to create much, much smarter computing systems that at least match human cognitive abilities like learning, reasoning, planning, creating, and remembering. information. .

It’s a sensible goal for a tech giant eager to shape the future of computing, attract top research talent and keep restless shareholders happy. But for you and me, this isn’t likely to mean that a hyper-intelligent robot will be offering advice through your smart glasses anytime soon.

This is because current AI, although exciting for computer scientists and much of the general public, has not yet been a revolution. He still has a hard time distinguishing concrete facts from fantasies. Even so, it’s still miles ahead of AGI, which exists primarily as a realm of research and speculation.

But it is Zuckerberg’s aspiration.

“It has become clear that the next generation of services requires building complete general intelligence,” Zuckerberg said in a post on the Meta’s Threads social network. “Building the best AI assistants (AI for creators, AI for enterprises, and more) needs advances in all areas of AI, from reasoning to planning, coding, memory, and other cognitive skills.”

Read more: AI chatbots are here to stay. Find out how they can work for you

And he means it, saying that by the end of the year, Meta will have 350,000 high-end Nvidia GPUs, the top-tier AI processors, which cost around $30,000 each. Adding other GPUs will yield processing power equivalent to 600,000 H100s, Zuckerberg said, dangling a big carrot in front of AI researchers’ eyes.

Giving a boost to his effort to usher in a metaverse that combines the real and computer-generated world, he said wearable devices like Meta’s Ray-Ban smart glasses could be an ideal interface to allow AI to see what that you see and help you navigate reality.

Today’s AI, best exemplified by large language models (LLMs) such as OpenAI’s GPT, Google’s Gemini, Anthropic’s Claude, and Facebook’s Llama, detects relationships between words in large expanses of text on the Internet, such as forum posts, books and news articles. The result is generative AI that can generate answers to many questions, style your job application answers to sound more formal, and more. But while LLM answers often seem plausible, these AI systems don’t really know anything.

What exactly is artificial general intelligence?

AGI, on the other hand, is more like your brain, broadly speaking. And with the constant progress of computing, there is a chance that if AGI is ever achieved, it will lead to superhuman intelligence later.

In an interview with The Verge, Zuckerberg didn’t offer any quick definition of AGI. “You can argue whether general intelligence is similar to human-level intelligence, or is like human-plus, or is some superintelligence from the distant future,” he said. “But for me, the important part is actually the breadth of it, which is that intelligence has all these different capabilities where you have to be able to reason and have intuition.”

OpenAI and Google’s DeepMind are among those pursuing AGI. Zuckerberg hopes Meta will be the company to deliver it.

To make that happen, he merged the company’s two AI research teams, the old school’s FAIR effort and the newer generative AI team building Llama.

How close are we to achieving AGI?

Opinions vary on whether today’s LLMs, which burst into mainstream consciousness with the arrival of OpenAI’s ChatGPT service based on its GPT model, are a step towards AGI. Microsoft researchers, in a 2023 paper, concluded yes.

“Given the breadth and depth of GPT-4’s capabilities, we believe it could reasonably be considered an early (although still incomplete) version of an artificial general intelligence system,” the researchers concluded.

However, there are many skeptics, including critics like Emily Bender of the University of Washington, who deride LLMs as mere “stochastic parrots” that regurgitate information somewhat randomly based on statistical patterns in their training data.

Others, including Facebook researcher and AI pioneer Yann LeCun, have argued that a sufficiently sophisticated training process can effectively build a representation of the real world in an AI model. In fact, some notable skills emerged from text-trained LLMs.

Researchers are now trying to advance AI with richer training data, a direction that leads toward a “world model” that captures our environment in much more depth.

LLMs train on text, but Google’s Gemini and other new “multimodal” AI models also train on video, photos, audio, and other data sources. Tesla CEO Elon Musk believes his company’s humanoid robots will collect useful video training data as they navigate the real world.

But more complete training data only goes so far if the same basic problems plague AIs, such as producing actions based on interpretations of a situation that are statistically plausible but not actually correct. Many researchers believe that simply improving current AI will be insufficient to achieve AGI.

And then there’s the thorny question of whether it’s even a good idea. Regulators, ethicists and computer scientists are debating the issue, but it is a highly speculative area and there is no consensus on how to control AGI-enabled machines or how to at least align them with the well-being of humanity.

Zuckerberg insists that Meta’s AI effort is proceeding cautiously, even with Llama 3 LLM having already begun training. “We have an interesting roadmap of future models that we will also continue to train responsibly and safely,” he said.

That’s good to hear, given the potential power and importance of AI. Judging by the struggles humanity has had with privacy erosion, social media misinformation, identity theft, and other technological problems, perhaps we should be grateful that Zuckerberg gave us at least a few years’ warning. advance notice of Meta’s AGI plans.

Editor’s note: CNET uses an artificial intelligence engine to help create some stories. For more information, check out this post.

Leave a Reply

Your email address will not be published. Required fields are marked *