OpenAI released its hotly-anticipated GPT-4 on Tuesday, providing a 98-page “technical report” on the latest iteration of its large language model (LLM).
But despite the lengthy documentation and the company’s not-for-profit roots, OpenAI has revealed extremely little information about how its latest AI actually works — which has experts worried, Venture Beat reports.
OpenAI, however, claims it had good reason to play its cards close to the chest.
“Given both the competitive landscape and the safety implications of large-scale models like GPT-4,” reads the paper, “this report contains no further details about the architecture (including model size), hardware, training compute, dataset construction, training method, or similar.”
In other words: this is OpenAI’s Krabby Patty formula, and they won’t be offering up the recipe anytime soon.
According to Lightning AI CEO William Falcon, an AI researcher who previously worked under Meta’s Chief AI Scientist Yann LeCun, OpenAI’s refusal to cough up their secret GPT-4 recipe is a precarious move.
OpenAI is “basically saying, it’s cool, just do your thing, we don’t care,” Falcon told Venture Beat, arguing that OpenAI has set a “bad precedent” for competing AI startups. “So 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.”
“These models can get super-dangerous very quickly, without people monitoring them,” he added. “And it’s just really hard to audit. It’s kind of like a bank that doesn’t belong to FINRA, like how are you supposed to regulate it?”
Show Your Work
Falcon makes an excellent point. It’s easy for anyone to say that they’re doing all of the right things to get to a certain outcome. But if you don’t actually show your work, outside regulation is pretty much impossible.
There’s also the reality that the existence of the paper is pretty misleading altogether. Though the OpenAI paper is called a technical report, it doesn’t exactly contain much technical information. And that, says Falcon, does everyone a disservice.
“You’re masquerading as research,” the CEO told Venture Beat. “That’s the problem.”
In any case, though, the technical report made one thing clear: that despite its name, OpenAI’s doors are firmly shut — and that’s unlikely to change.