(2024-04-19) Evans Looking For Ai Usecases

Benedict Evans: Looking for AI (GenAI) use-cases. ....in the last 18 months, as I’ve experimented with ChatGPT, Gemini, Claude and all the other chatbots that have sprouted up: ‘this is amazing, but I don’t have that use-case’.

The one really big use-case that took off in 2023 was writing code, but I don’t write code. People use it for brainstorming, and making lists and sorting ideas, but again, I don’t do that.

This wouldn’t matter much (‘man says new tech isn’t for him!’), except that a lot of people in tech look at ChatGPT and LLMs and see a step change in generalisation, towards something that can be universal

as these models get better and become multi-modal, the really transformative thesis is that one model can do ‘any’ use-case without anyone having to write the software for that task in particular.

It seems to me, though, that there are two kinds of problem with this thesis.

The narrow problem, and perhaps the ‘weak’ problem, is that these models aren’t quite good enough, yet.

They’re now very good at making things that look right, and for some use-cases this is what you want, but for others, ‘looks right’ is different to ‘right’.

The deeper problem, I think, is that no matter how good the tech is, you have to think of the use-case. You have to see it. You have to notice. (killer app)

reminds me a little of the early days of Google, when we were so used to hand-crafting our solutions to problems that it took time to realise that you could ‘just Google that’.

However, the other part of this pattern is that it’s not the user’s job to work out how a new tool is useful.

Dan Bricklin, and in principle all software, had three steps: he had to realise that you could put a spreadsheet into software, then he had to design and code it (and get that right), and then he had to go out and tell accountants why this was great.

In that case he had perfect product-market fit almost immediately and the product sold itself, but this is very rare.

The great recurring fallacy in productivity software startups is that you can sell bottom-up without a sales force, because the users will see it and want it

Hence, one hypothesis today might be that generative AI could remove or minimise Dan Bricklin’s work actually to build the product, but you still need to realise that you could do this, make something tangible that expresses that, and then go out and tell people

you probably don’t want to give ChatGPT to Dwight or Big Keith from The Office and tell them to use it for invoicing, anymore than you tell them to use Excel instead of SAP.

I often compared the last wave of machine learning to automated interns

machine learning’s breakthrough was over a decade ago now, and yet we are still inventing new use-cases for it

You could propose the current wave of generative AI as giving us another set of interns, that can make things as well as recognise them, and, again, we need to work out what.

we would still have an orders of magnitude change in how much can be automated, and how many use-cases can be found for LLMs, but they still need to be found and built one by one.

That would make LLMs the new SQL, not the new HAL9000.