Palantir CEO Alex Karp gives his candid thoughts on the hottest buzzwords in the AI industry. Speaking on the TBPN podcast in conjunction with Palantir’s AIPCon 10 conference, Karp likened “tokenmaxxing,” the compulsive overuse of AI tokens, to porn addiction. What he meant was simple. Many companies are running out of tokens that look like productivity but actually provide nothing.Karp told organizers that Palantir has built internal tools to wean companies away from this practice. Although the name doesn’t fit the air, he described it as “stop masturbating, stop masturbating.” “People just sit there all day like they’re addicted to porn,” he says. Despite the organizers’ attempts to steer him to the analogous AI side, he kept returning to the comparison.Tokens are building blocks of larger language models that divide words into numerical units. A single token is about three-quarters of a word. AI providers typically charge fees based on the number of tokens consumed and the model chosen. Over the past few weeks, some in Silicon Valley have railed against tokenmaxxing, a culture that supports near-unlimited use of AI.
Tokenmaxxing backlash is AI productivity Mathematics stops adding
Companies like Meta and Amazon have reportedly built internal scoreboards to track token consumption, treating raw usage as a proxy for productivity. Since then, most have made cuts. Uber Chief Operating Officer Andrew McDonald said the ride-hailing company has struggled to translate AI-related price increases into real profits. Until about two weeks ago, Karp said, executives felt they wouldn’t look foolish for questioning the AI publicly, even though they privately acknowledged it wasn’t fully functional.His chief technology officer, Shyam Sankar, made a more pointed statement during a recent earnings call, calling Palantir a “no-slop zone.” More tokens means more slop, Sankar said.
Why Karp thinks LLMs will hit a wall with difficult domain-specific problems
Karp did not completely deny AI. He said it was real and could be useful for narrow tasks, such as writing a report on China’s GDP growth. The problems start with more difficult sector-specific problems, such as finding cheaper, legal and ethical ways to drill for oil and gas. These, he argued, require precise and ongoing processes that LLM can enhance but not replace.Of course, much of it also doubles as a sales pitch. Karp told potential clients to first spend a few days with frontier companies like OpenAI and Anthropic, then call them when they’re done. Because clients usually come back “yelling.”
