Tesla CEO Elon Musk has predicted that the AI race will advance manifold by the next year, or by 2026, and will outsmart the “smartest human”.
"If you define AGI by if it’s smarter than human, I think it's possible next year or like within two years.” He adds that for AI to become smarter than machine-augmented or all humans working together, who will also use computers to augment output, it’s “probably five years away."
Musk says there's too much advancement that is happening in the AI space, with the technology moving at the fastest pace. “I have seen a lot of technology. There's never a week that goes by without some new announcement,” Musk says during a podcast ‘In Good Company’ with Norway wealth fund CEO Nicolai Tangen.
Musk says if one looks at computer hardware dedicated to AI, its sophistication is increasing at least by a factor of 10 every year. “So when you combine the hardware, the order of magnitude increases every nine months, and many many software breakthroughs..."
“I think we'll have an AI that is smarter than any one human, probably, around the next year, and then the total computing of AI will probably exceed humans in five years.”
On roadblocks in the AI race currently, the Tesla CEO says: "Chip constraint, hardware deployment, and data (are the main constraints). Last year, it was chip supply, this year it's starting to transition into voltage transformer supply. My sort of very niche joke is “Transformers for Transformers”. Next, in a year or two, the electricity availability will be the main constraint."
The billionaire, whose xAI company competes with the likes of OpenAI and Google, says many of the smartest humans are working on AI because it's the fastest-moving field. “A lot of the best talents and smartest are going into AIs. Along with that, we see algorithmic breakthroughs. Then you start hitting the wall with data problems."
Explaning the data constraints in training AI models, Musk says one can fit all the books written ever, in compressed form, in one hard drive in one computer. “All the books ever written, in any language, by all the humans, sound like a lot, but it's actually in a smaller number of training tokens. It's not enough. So then you start looking at all the videos ever created. All the podcasts, all the everything...and you start running out of data there."
On the biggest challenge his company xAI is facing in training the AI models, Musk says xAI is still relatively new, so the company is currently training the Grok 2 model, which he claimed will be better than GPT-4. “We hope to complete that in May. The next step should be for Grok 3, which should beat GPT 5 or beyond."
To solve data issues, says Musk, one has to either create “synthetic data” or use “real-world video”. These two are supposed to have unlimited data, and Tesla has a pretty good advantage over real-world video.
Leave a Comment
Your email address will not be published. Required field are marked*