Spent an inordinate amount of time today reading and listening to AI-sceptic views from Ed Zitron (on his Better Offline podcast, plus articles like The Case Against Generative AI), The Verge with its Vergecast pod, with the Episode AI can’t even turn on the lights (you may want to skip the whole baby-tech portion at the beginning) and various associated links from these main sources (Colton Voege’s post No, AI is not Making Engineers 10x as Productive is a great one).
They align with (and are massively better researched, considered and thought-through than) my own thoughts - more feelings - on the matter, which are basically sceptical and uneasy about using such tools for very much at all.
Fundamentally, it boils down to the fact that these tools can be good, but there is no way that any gains they may produce can justify enterprises (or advertisers) spending enough money on the LLM providers to recoup the simply mind-boggling costs that they are incurring but also forcing through to “beat” all the others in gaining some kind of edge through which they will survive the bursting of the bubble.I have subscribed to one additional, smaller, model - Le Chat, by Mistral - along with the standard Gemini model that comes with my Google One subscription, and another cat, called Lumo, part of another data and mail provider, Proton. They’ve been good when researching vague topics, not great when researching something very specific to the world of trombones - and not earth-shatteringly good.
The bigger, global arguments about their being an investment sink and massive consumers of power and water are much more important to us all, and I will, where I can, try to ensure that my employer deploys a good amount of scepticism and due-diligence when looking to implement AI into its systems. From what I’ve heard thus far, we’re doing well on that front, not leaping into AI too madly
