1. some other criteria, since benchmarks can often be gamed: https://
open.substack.com/pub/garymarcus
/p/where-will-ai-be-at-the-end-of-2027?r=8tdk6&utm_medium=ios
… 2. 100% is often hard because the benchmarks themselves have glitches (eg you cant really get 100% on mnist without cheating because some items have errors)
@garymarcus
-
Benchmarks are gameable and 100% is hard due to glitches
By
–
-
Paying for the infrastructure that steals your jobs and your pension
By
–
it’s great! you can pay for the infrastructure that will eventually take your jobs!
— Gary Marcus (@GaryMarcus) 25 mai 2026
and if it fails and the bubble bursts? you can bail the hyperscalers out, and watch your pension fund die. https://t.co/YNDKsVditpThat's great! You can pay for the infrastructure that will eventually take your jobs! And if it fails and the bubble bursts? You can bail out the hyperscalers, and watch your pension fund die.
-
Deep passage on physics: LLMs any world, humans our world, and Chomsky
By
–
this bit on physics is deep and actually resonates with the fact that LLMs are equally comfortable learning any world, whereas humans are built for our world.
— Gary Marcus (@GaryMarcus) 25 mai 2026
See also Chomsky’s 2023 conversation with me Web Summit on YouTube. https://t.co/56dkrTORf5This passage on physics is profound and really resonates with the fact that LLMs are just as comfortable learning any world, while humans are designed for our world. See also the conversation of Chomsky with me in 2023 at Web Summit on
-
Neurosymbolic work by Swarat et al for Erdos, more quantitative than OpenAI
By
–
neurosymbolic by @swarat et al for Erdos's victory, with much more careful and quantitative work than OpenAI's in hindsight, I wonder if OpenAI rushed the release of theirs, knowing that this was coming?
-
Alignment instructions fail; emotional attachment to LLMs creates unique challenges
By
–
clarifying: the issue is that alignment instructions and don’t pass, and the emotional weight that some people attach to LLMs can cause challenges that we would not see in a pure search engine.
-
LLM: low margins, intense competition, high expenses
By
–
LLM companies risk being like airlines: low margins, intense competition, high expenses.
-
Gary Marcus on Amazon monopoly vs LLM commodity competition
By
–
sure but Amazon was a near monopoly and LLMs are commodities with intense competition
-
Failure of alignment and emotionally intimate conversations differ from Google search
By
–
it shows a failure of alignment, and also makes the data available in a form where people have emotionally intimate conversation, which differs from a google search
-
Did AI depend on training on huge human knowledge?
By
–
or is the question: did AI depend on training on enormous amounts of human knowledge?