I don't buy it that AI will widen social divides. The evidence points the other way. LLMs are mostly useful for beginners, and they are basically a fast lane to knowledge. It's never been easier to start from nothing, absorb lots of synthesized knowledge, and go beyond what an LLM can do - essentially becoming "the best" according to the article. The path to being "the best" in any discipline is not decades today, it is years.<p>Personally, I've picked up so many new skills in the last five years thanks to the LLMs. I now do my own car maintenance (fully), I'm restoring my home, and handling years of programmer health neglect. Each of these would have taken me either years to learn, or at least months with professional supervision. Thanks to LLMs, I basically have a direct line to a somewhat knowledgeable "person" that can answer all my questions immediately, and it takes days to learn how to be somewhat good at anything. I'm not saying I've become an expert car mechanic, builder, or personal health expert, but I have become <i>functionally</i> good. I can look at my work and say "I've actually done this better than the contractor I hired 2 years ago!", etc.<p>It's true that there is a divide between "the best" and "the rest", but the divide is that "the best" don't benefit much from LLMs. At work where I am a senior SWE, there is nothing the current LLMs (including GitHub Copilot) can help me with. But on the weekends, when I'm learning a fun new programming language <i>as a beginner</i>, I can get up to speed in a few hours. That is the different effects LLMs have on "the best" and "the rest".<p>If LLMs did not exist, and one needed to have really extensive domain expertise, plus ML, tensorflow, and python skills - then I would agree with The Economist that AI benefits technologists more. But LLMs exist, are widely available, and also are probably the main application of ML today. So I think the article misses the point very much.