OpenAI unveiled GPT-4 on Tuesday, but the accompanying 98-page technical report left out crucial details about the model's inner workings, prompting concerns from AI researchers and industry leaders about the potential risks of such secrecy.
The report, which OpenAI itself labels a "technical report," explicitly states that it contains "no further details about the architecture (including model size), hardware, training compute, dataset construction, training method, or similar." The company cites "the competitive landscape and the safety implications of large-scale models like GPT-4" as reasons for withholding this information.
William Falcon, CEO of Lightning AI and a former researcher under Meta's Chief AI Scientist Yann LeCun, criticized OpenAI's decision in an interview with Venture Beat. Falcon argued that by keeping the details under wraps, OpenAI is setting a "bad precedent" for other AI companies, potentially discouraging open-source practices across the industry.
"You are going to have all these companies who are not going to be incentivized anymore to make things open-source, to tell people what they're doing," Falcon said. He warned that without external oversight, these models "can get super-dangerous very quickly," and compared the situation to a bank that doesn't belong to FINRA, making regulation nearly impossible.
The critique extends beyond the lack of transparency. Falcon accused OpenAI of "masquerading as research," noting that the paper, despite its title, contains little technical substance. This, he says, does a disservice to the broader research community and hampers efforts to audit and regulate such powerful AI systems.
OpenAI's decision to withhold details is not unprecedented in the competitive AI landscape, but it marks a notable shift for a company that began as a non-profit. The move has sparked a broader debate about the balance between commercial interests and the public's right to know about technologies that could have significant societal impacts.
The release of GPT-4 itself has been met with both excitement and caution. The model has demonstrated impressive performance on various tests and exams, but the lack of transparency about its training data and architecture raises questions about its potential biases and limitations.
As AI continues to advance at a rapid pace, the call for greater openness and accountability is likely to grow louder. For now, OpenAI's technical report serves as a reminder that in the world of AI, what is left unsaid can be just as important as what is disclosed.
