Alexander Embiricos, who leads the development of the Codex programming agent at OpenAI, said that human speed is an «underestimated limiting factor» for creating artificial general intelligence (AGI).
Alexander Embiricos, who leads the development of the Codex programming agent at OpenAI, said that human speed is an «underestimated limiting factor» for creating artificial general intelligence (AGI).
Many large companies are currently aiming to develop a version of AI that can reason as well or better than humans. According to Embiricos, this is hindered by «human typing speed» or «human multitasking speed when writing prompts,» Business Insider reports .
«You can have an agent that’s monitoring all your work, but if you don’t have an agent that’s also reviewing their work, then you still face the problem of: can you review all this code?» the Codex lead told Lenny’s Podcast.
He believes that we need to free people from having to write prompts and check the AI’s performance, because it slows down the process. «If we can redesign systems so that the agent is helpful by default, we will start to unlock the potential,» he said.
Embiricos noted that there is no simple path to a fully automated workflow—each use case will require its own approach—but he expects progress to be made in this direction soon.
He stated that somewhere between the moment when the first users start to see productivity gains and the moment when tech giants can fully automate processes using artificial intelligence agents, AGI will emerge.
Previously, Meta AI researcher Yann LeCunh said that large language models (LLMs) are useful but limited. In his opinion, this direction will not be able to be developed to the level of artificial general intelligence (AGI).
