OpenAI Employee Claims AGI Has Been Achieved

An OpenAI staff member has sparked debate by suggesting that the company has reached artificial general intelligence (AGI), albeit with caveats.

OpenAI Staffer’s Bold Statement

Shortly after OpenAI’s release of its latest AI model, O1, Vahid Kazemi, a member of the company’s technical team, made a striking claim on X.

“In my opinion,” Kazemi stated, “we have already achieved AGI and it’s even more clear with O1.”

However, his assertion came with a significant qualifier.

“We have not achieved ‘better than any human at any task,'” Kazemi clarified, “but what we have is ‘better than most humans at most tasks.'”

Redefining AGI

Kazemi’s comments have drawn attention for relying on a broad and unconventional definition of AGI. Rather than asserting the AI surpasses human expertise in specialised tasks, he emphasised its versatility, suggesting it can handle a wider array of tasks than any human could manage.

Critics argue this interpretation stretches the conventional understanding of AGI, which is typically associated with human-level performance across all domains.

The Nature of LLMs

Kazemi also addressed the ongoing debate about whether large language models (LLMs) are simply mechanistic systems that “follow a recipe.”

“Some say LLMs only know how to follow a recipe,” he wrote. “Firstly, no one can really explain what a trillion parameter deep neural net can learn. But even if you believe that, the whole scientific method can be summarised as a recipe: observe, hypothesise, and verify.”

This defence aligns with OpenAI’s philosophy that scaling up data and computational power in machine learning systems could eventually lead to true human-level intelligence.

“Good scientists can produce better hypothesis [sic] based on their intuition, but that intuition itself was built by many trial and errors,” Kazemi explained. “There’s nothing that can’t be learned with examples.”

Business Implications

Kazemi’s remarks come amid recent news that OpenAI removed “AGI” from the terms of its partnership with Microsoft, raising questions about the business context of his claim.

For now, no AI system has proven capable of competing with humans in the workforce on a truly general level. Should that day come, assertions like Kazemi’s may warrant closer scrutiny.