OpenAI co-founder: "AI training on data from the Internet is nearing completion, then the technology will make decisions on its own"
Artificial intelligence will likely soon become more autonomous than it is now.
Artificial intelligence will likely soon become more autonomous than it is now.
Artificial intelligence will likely soon become more autonomous than it is now.
Ilya Sutzkever, co-founder of OpenAI, which developed the ChatGPT AI model, shared his vision for the near future in the development of artificial intelligence.
He believes that the era of pre-training models on data from the Internet is coming to an end, as this data has its end. “The Internet is one, there will be no new data,” he said, calling data “the fuel for AI training.”
Sutzkever believes that the next generation of models will have “agency” — the ability to make decisions and perform tasks on their own. This way, these models will be able to reason, rather than simply match patterns based on what they’ve seen before. And the more the system reasoned, the less predictable it became. “It became like unpredictable and advanced AI chess for the best chess players,” Sutzkever says.
He also drew a parallel between AI scaling and evolutionary biology, noting that hominids (human ancestors) had a different brain-to-body mass ratio than other mammals. He argues that just as evolution found a new scaling pattern for the hominid brain, AI can find new approaches to scaling that go beyond current pre-training.


