I always appreciate our weekly Crankypants Take on LLMs.
> AI is not a triumph of elegant design, but a brute-force workaround
You can read and understand Attention Is All You Need in one hour, and then (after just scaling out by a few billion) a computer talks to you like a human. Pretty elegant, if you ask me.
> The web was supposed to evolve into semantically structured, linked, machine-readable data that would enable amazing opportunities.
I missed that memo. The web was, and forever shall be, a giant, unstructured, beautiful mess. In fact, LLMs show just how hopeless the semantic web approach was. Yes, it's useful to attach metadata to objects, but you will still need massive layering and recursion to derive higher-order, non-trivial information.
This entire article is someone unable to let go of an old idea that Did Not Work.
> AI is not a triumph of elegant design, but a brute-force workaround
I think the author is on to something here, but does not realize it, or applies the think to the wrong problem. The issue isn't the web, search was good enough and it's a pretty big problem to solve. We need to go smaller. Applying AI to customer service, internal processes and bureaucracy, among other things, is an inelegant brute-force approach to not just fixing your shit.
The majority of customer service could be fixed by having better self-service, designing better UIs, writing better manuals, having better monitoring, better processes, better trained staff, not insisting on pumping stock numbers and actually caring about your product. AI will never fix the issues your customers are having, they'll just make customer service cheaper and worse because just like real humans the content and levers they need aren't available.
Same for much of the bureaucracy in companies and governments. Rather than having AI attempt to fill out forms or judge if someone is entitled to a pension, insurance pay out or what have you. Take the time to fix your damn processes and built self-service portals that actually works (some of those may have some sort of AI on the backend for things like scanning documents).
Forget the semantic web thing, it never worked, to much data, and AI generated content certainly isn't making the problem easier. Let's work on that at some other time. Still the author is correct, LLMs are a brute-force workaround, regardless of how elegant the design may be. They are applied to a host of problem that should just be eliminated, not hidden and stupefied by a prediction engine.
> AI is not a triumph of elegant design, but a brute-force workaround
You can read and understand Attention Is All You Need in one hour, and then (after just scaling out by a few billion) a computer talks to you like a human. Pretty elegant, if you ask me.
> The web was supposed to evolve into semantically structured, linked, machine-readable data that would enable amazing opportunities.
I missed that memo. The web was, and forever shall be, a giant, unstructured, beautiful mess. In fact, LLMs show just how hopeless the semantic web approach was. Yes, it's useful to attach metadata to objects, but you will still need massive layering and recursion to derive higher-order, non-trivial information.
This entire article is someone unable to let go of an old idea that Did Not Work.