Well this story is just a pure delight. In 2017, the Chinese said that by 2030, they were going to be the world champions of AI. So in 2022, the U.S. put export controls on the fancy computer chips, especially Nvidia chips, that AI needed. Then in 2025, the Chinese announced an AI entity called DeepSeek that used, in part, outdated Nvidia chips but mostly inventive software. DeepSeek works as well as or better for AI’s LLMs, the large language models like ChatGPT. It uses 2,048 of those old Nvidia chips, and costs $5.6 million. An LLM built by Meta used 16,000 new Nvidia chips, and cost $60 million.
I’d say — quoting Pappy O’Daniel — the U.S. got its behind kicked. Which as a U.S. citizen, I don’t find delightful. The delight is this: years ago I was interviewing a famous old Cold Warrior, Sid Drell, about the US/USSR Apache standoff over missile defense systems. He said, “What you want to be very careful of, is not the make the fallacy of the last move.”
I didn’t, and still don’t, know the origin of that phrase. The earliest occurance of it I could find was by Herb York, another famous old Cold Warrior, writing about missile defense in 1969. York is talking about the US learning about a USSR system to protect itself against incoming US missiles, so the US invented another system to arm its missiles with lots of warheads aimed at lots of Russian targets, so the Russians built its own multi-warhead missiles. Etcetera, far into the night. One side tries to sneak-launch a warhead-carrying missile; the other side puts up a satellite that detects missile plumes’ infrared radiation the minute the missile launches; the first side might then hide the plumes’ infrared with a nuclear explosion whose own infrared would swamp the plumes’. Et cetera, far into the night, emphasis on “night.”
I like the way Drell described it to me. “The ABM [anti-missile] problem was a human problem. . . It was a technical problem to put a man on the moon. And that’s because the moon didn’t mind being landed on. It couldn’t put out decoy moons, it couldn’t put out its lights, it couldn’t run away. . . But the fact is, ABM is a human problem. Because if we built a system, they’ll build counter-measures, and we’ll build counter-counter-measures.” That’s when he said, “And what you want to be very careful of is, not the make the fallacy of the last move.”
Drell really nails the fallacy of the last move, doesn’t he, the completely human and astoundingly dumb idea of thinking that because you smack somebody down, they won’t get up and smack you down right back. They declare they’re going to beat you at AI, you limit their access to fancy chips, they do without and write fancier programs. Covid-19 hits an immunologically-naive population and becomes a pandemic; the population develops a vaccine against it; covid-19 changes so the vaccine doesn’t work; the vaccine-makers make a new vaccine that works. The blue team elects a Democrat, the red team elects a Republican, the blue team elects a Democrat, the red team re-elects the Republican who’s rooting out every blue molecule in the country. None of these fights have finished, none of these moves are the last ones. I’m so excited for the next ones.
(True last moves are pretty rare; the only ones I know of for sure are gravity, water, squirrels, and death.)
So if last moves are a fallacy, then what? Drell spent his life working out the answer, which he wrote in many articles and books, of which I’ve read only a few and which I didn’t necessarily understand. His answer had to do with treaties that moved the attention of the warring parties from technical problems to human ones. Smart guy.
I don’t mean to say that treaties are the only way to solve the fallacy of the last move — but maybe if not actual treaties, then the principle of them? Like, guys, nobody’s happy here, how about if we work out some kind of chips/software agreement so that neither of us loses honor or money, and the world gets cheaper and better AI*? Nobody’s going to listen to me, are they.
So my point is, don’t be thinking that just because you’ve got an effective countermeasure, the other guy isn’t going to counter-countermeasure you right back. The last move is a fallacy. It was a fallacy when Reagan funded the Strategic Defense Initiative, it was a fallacy when Bush tried to revive it, it’s still a fallacy when Trump tries to install an Iron Dome. He/she/they/it is going to kick your behind. And you’ll kick back, far into the night. Emphasis on “night.”
____________
*I know. It’s not this simple. The Chinese are apparently more interested in AI that can help with manufacturing, while the US is more, as this article says, “consumer-facing.” That is, maybe we don’t need a treaty, maybe we’ll just move into separate niches. That would work too.
___________
Photo of Hou Yifan at the 2016 Chess Olympiad by Andreas Kontokanis, via Wikimedia Creative Commons.
Photo of Sid Drell looking exactly the way he looked at people, by Jim Harrison for the Heintz Awards; and I don’t know if this is Creative Commons or not but if Jim or Heintz wants me to take it down, I will.
Classic. My older brother liked putting me in my place, especially when we competed in games. He would say “David, I am ten times better than anything you say.” I would reply “well, I’m five times better than you”, to which he would promptly assert “that makes me fifty times better than you!” On and on it would go, and I would be defeated each time. Then, one day, I found the way to beat him. After he again challenged me with being ten times better than anything I say, I announced “well, I am negative five times better than you!” He stood there for just a moment, shocked that I had finally bested him, and then he punched me, ten times harder than I could have punched him.
I LOVE that story! It just goes to show, doesn’t it.