Ben Thompson’s latest piece, Enterprise Philosophy and The First Wave of AI is an interesting (and long) take on how he imagines AI being built into the real world.

It’s worth reading, but I think an item from the summary stands out to me:

My core contention here, however, is that AI truly is a new way of computing, and that means the better analogies are to computing itself. Transformers are the transistor, and mainframes are today’s models. The GUI is, arguably, still TBD.

(Emphasis mine).

I’m not convinced (yet?). I still have not seen an implementation that demonstrates substantially different capabilities than previous tools, with two exceptions:

• Visual media manipulation (generate images, videos, remove objects)
• Slop text generation

While visual media manipulation is impressive (I use the latest features in Lightroom all the time), it still feels augmentative more than generative so far as it’s impact. I’m probably under-appreciating the degree to which “generate an image of X for me” will be impactful, but it feels like the outcome is far more likely to be more images generated than it is replace the human who builds our UI¹.

Same with text generation: How many people are actively writing slop right now, who are in danger of being replaced by a computer doing it? More than zero, I’d wager, but this doesn’t seem society-transforming (maybe in its annoyance, but that’s something else).

Neither of these feel like “a new way of computing” to me. So perhaps this AI future is in “Agents”. But where is the demonstration that this technology can lead to these agents existing, let alone working well enough to be considered a paradigm shift?

The improvement curve of these LLMs seems to be flattening, and having the LLM run it’s own multi-shot answer to each question (like o-1 is doing) might be more effective, but it’s hardly a game-changer, IMO. So where are the orders-of-magnitude improvement going to come from?

I’m open to the idea it could happen, but I’m not seeing it as remotely inevitable along the existing axes of improvement.

Maybe Ben’s right, and this really is a generational transformation, in which I’ll struggle to adapt, if I adapt at all, and that I’m too entrenched in existing ways of computing to think outside the box.

Or, maybe this is more like grocery store self-checkout, where every large organization will, behind the dream of free labor, spend considerably more money than they could ever hope to save building complex, obnoxious contraptions that fail to do the job as well as a 17-year-old clerk.

MidJourney is every going to be, IMO.