chrismdp

Great thoughts. I too came back from vacation this year to find that I wasn't nearly as interested in AI doing certain things for me (like writing) and much more interested in finding the areas it's properly good at.

At the moment I'm getting most value out of: - Checking integrity of data through custom scripting - Controlling my smart home stuff via quick commands - Reconciling account software with my bank and uploaded / emails receipts - Researching info across disparate sources - Managing pipelines of work (eg sales cycles or content schedules) reminding me when I need to do something - Parsing interesting articles / videos and pulling out the essentials - Reworking text I've written to work across multiple social accounts - Asking me great questions about whether I'm on track with what I said I wanted to do - Pinging me when an email looks important so I don't have to continually check it. - Coding on projects that I want to make rapid progress on for work (not tinkering)

This list is quite different to the one I'd have written a year ago - that would have mostly been about writing, information retrieval and coding.

The most obvious use cases are I think the least useful ones. Diffusion of AI will take time!

It is nothing new - and that makes it predictable and scary.

> The final puzzle piece

...to what? That's what I'm worried about!

100%. You can't and shouldn't wrap every interaction. We need a new approach.

I think the building blocks of the most impressive experiences will come from choosing the exact right point to involve an LLM, the orchestration of the component pieces, and the user experience.

That's certainly what I found in games. The games which felt magic to play were never the ones with the best hand rolled engine.

The tools aren't there yet to ignore prompts, and you'll always need to drop down to raw prompting sometimes. I'm looking forward to a future where wrangling prompts is only needed for 1% of my system.

It's always true that you need to drop down a level of abstraction in order to extract the ultimate performance. (eg I wrote a decent-sized game + engine entirely in C about 10 years ago and played with SIMD vectors to optimise the render loop)

However, I think the vast majority of use cases will not require this level of control, and we will abandon prompts once the tools improve.

Langchain and DSPY are also not there for me either - I think the whole idea of prompting + evals needs a rethink.

(full disclaimer: I'm working on such a tool right now!)