Palantir CEO Alex Karp doesn’t buy into the “tokenmaxxing” hype.
Never one to mince words, Karp likened the urge to use AI in such an insatiable way to watching porn.
“In fact, internally we call this de-mastication, like stopping masturbation,” Karp said in a live interview with TBPN on the sidelines of Palantir’s AIP Con 10, regarding Palantir’s token tracking. “Sure, people are like porn addicts and they just sit there all day.”
Karp’s comments echo what Palantir Chief Technology Officer Shyam Sankar told analysts on an earnings call last month about the data analytics and technology company calling itself the “Northrop Zone.” Sankar said companies need to understand that cheaper AI alone won’t create more value unless they have systems like Palantir’s AIP (artificial intelligence platform) that can establish AI models.
“More tokens means more slop,” Sankar said. “And the more awareness a product has, the more you need a system to prevent economic harm so you can leverage the economic value.”
Tokens are building blocks of a larger language model that breaks down words into numerical units. One token is about 3/4 of a word. AI companies and AI model providers often charge fees based on the number of tokens consumed and the models used.
In recent weeks, parts of Silicon Valley and the tech community have railed against “token maxing,” a culture that favors the rise of AI agents and the near-free use of AI to match their capabilities.
Uber Chief Operating Officer Andrew MacDonald highlighted these concerns, saying the ride-hailing company is struggling to see the link between rising AI prices and meaningful benefits such as increased productivity. In relation to MacDonald’s comments, Karp said that until recently it probably wasn’t considered wise to publicly raise questions about AI.
“When we first met, it was like an AI, maybe real,” Karp said of the first time he met the TBPN crew. “So, for some reason, up until about two weeks ago, I think something terrible happened. Which is true, but somehow it’s not working. But I’m not allowed to say that publicly, because it makes me look stupid.”
Karp said that while people recognize that AI is real, many of the questions surrounding the technology come down to a matter of “preference,” including whether competitors will try to build something like Palantir’s ontology.
“All of these things can be scaled up to be very valuable, but they’re going to be largely commercialized. But you can’t scale up what are the business problems that you want and need to solve,” he said.
There are some problems that AI models can solve very well, Karp says. He gave an example of a prompt such as “I would like to write a report on China’s GDP growth.”
He said there are more complex dilemmas that AI alone cannot solve.
“I want to understand ways specific to oil and gas drilling that are legal and ethical and reduce production costs. I want to change the supply chain in my industry, whether it’s the military, whether it’s box manufacturing, whether it’s automotive. These require real, precise, ongoing processes,” Karp said. “They are enhanced by large-scale language models; they are not replaced by large-scale language models.”