Message of 27 May, George HoltzGeorge HoltzOn May 24th, a blog was published, stating that in the course of the current software development, it is not possible to find a solution to the problemAI ProgrammingAgentPerhaps one of the most costly mistakes。
note: hotz was born in october 1989 and is known as "geohot", a world-renowned hacker, safety researcher and entrepreneur。
He is known for his early efforts to decipher iPhone and PlayStation 3 and is the founder of the auto-driving technology company Comma.ai, which has received considerable attention in terms of programming and artificial intelligence in the form of technocrats。
according to holtz, he has tested various models and tools for six months and has combined the work of tinygrad with the conclusion that they are suitable for fast prototypes but are not very good at dealing with the details of the quality of decisions。
IN BOVEN, HOLTZ WAS CONCERNED THAT LARGE ORGANIZATIONS WERE OVER-RELIANT ON THE AI TOOL, AND HE NOTED THAT THE AI MODEL COULD PRODUCE CODE QUICKLY, BUT THAT THE RESULTING SURFACE WAS BECOMING MORE AND MORE LIKE AND THE PROBLEM WAS MORE DIFFICULT TO IDENTIFY。
According to Holtz, it is difficult for less capable developers to see through these deficiencies and may eventually bring problematic codes into the formal system, accumulating high maintenance costs and hidden downside risks。
In technical judgement, Holtz has moved from early optimism to suspicion. He's now closer to Yann LeCun and Gary Marcus' point of view that the Big Language Model is not really intelligent。
In his view, such models were inherently complex statistical systems, and the main task was to imitate programming distribution rather than to really understand problems. That is why they can mimic the shape of the code, but they may not be able to stabilize the reasoning and modification in strange circumstances。
Contrary to Holtz, Andrey Capasi's attitude has recently clearly turned positive. In the autumn of 2025, he considered the smart system unreliable, but changed it after GPT-5.4 and Opus 4.6 were released, stating that the AI smart body had permanently changed the programming。