AI Companies Are Lying To Us

TL;DR: Sam Altman and Greg Brockman insist that super-intelligence will unlock “universal high income,” but their own words reveal a compute-for-everything scheme that replaces cash UBI with company-issued “points,” leaves trillionaires intact, and taxes nothing—an end-run around public AI dividends. ## The 90-Minute “Palace Intrigue” in One Room OpenAI’s last two remaining co-founders, Sam Altman (CEO) and Greg Brockman (President), sat down together for the first time on a single podcast—*Core Memory* with Ashlee Vance and Kylie Robinson—to mark the lab’s tenth birthday. The conversation quickly pivoted from nostalgia to the future they are still pitching: artificial general intelligence (AGI) followed quickly by artificial super-intelligence (ASI) that “adapts to humans” and floods the planet with cheap cognition. ## “People Who Really Know How to Use These Agents Will Become Trillionaires” Altman repeated a line he has tested in several recent talks: mastery of AI agents, not mere access, will be the new source of extreme wealth. When pressed on why super-intelligence still demands human “expertise,” he answered that the gap between “can do anything” and “knows what to ask for” will remain wide, creating a new class of trillion-dollar operators. ## Universal High Income, Not UBI—and Definitely Not a Tax The pair explicitly rejected permanent cash transfers funded by AI profits. Instead, Altman sketched a future where OpenAI (and presumably its peers) issues “compute credits” redeemable for tokens, cycles, or inference time. - “If compute is made accessible to everyone that could lead to the most extreme version of the American Dream,” Altman said. - Brockman added that their capped-profit vehicle has already earmarked “roughly 30 % of OpenAI equity” for a nonprofit fund whose charter is still being written. No mention of congressional tax legislation appears anywhere in the 90-minute interview. ## Compute Points as Currency: What You Can (and Can’t) Build Altman rattled off low-capital creative goods—movies, books, AI influencers—as examples of what a teenager could conjure with a modest stipend of GPU hours. The hosts asked how an individual would finance FDA trials for a newly designed drug. Brockman conceded that “dry-lab” discovery is only step one; wet labs, Phase III trials, and manufacturing still require “millions in funding,” leaving billion-dollar incumbents like Pfizer with their moats intact. ## Regulatory Capture and the Safety Narrative Both executives argued that iterative deployment—shipping ever-larger models and watching them break—*is* their safety strategy. Brockman said early internal fights were not over whether to be safe, but over “how to talk about safety without it becoming a power move.” The result: OpenAI now lobbies for light-touch, pre-deployment testing frameworks rather than revenue-linked liability or compute taxes. ## The Thirty-Percent Fig-Leaf OpenAI’s original 2015 nonprofit owns a minority share of the for-profit holding company. Altman reiterated that the board can “pull the plug” if humanity-level risk appears, but dodged questions about dividend flow, share-class voting power, or what happens once the capped-profit investor return hurdle ($100 × initial stake) is met. Critics note that 30 % is a ceiling, not a floor, and can be diluted in future funding rounds. ## “Extreme American Dream” or Compute Feudalism? Vance summarized the model: “So the state doesn’t tax the AI winners; the winners hand out tokens and keep the cash.” Altman laughed, calling it “the most inclusive form of abundance ever,” then pivoted to anecdotal stories of users who already use ChatGPT to force insurers to approve MRIs. No data on average user earnings or the distribution of compute vouchers was offered. ## What’s Next: GPT-5, Agents, and Robotics - GPT-5 will be “materially better” but not yet ASI, Brockman said. - A swarm of agent prototypes is running internally; external alpha “later this year.” - Robotics partnerships (Figure, 1X) will feed physical-world data back into the model loop, accelerating the path to “fully general” systems. ## Source [AI Companies Are Lying To Us – Gullible_Pen1074](https://youtu.be/NCKQL0op30E)