How to keep up with AI/ML as a full stack dev?

I'm in a similar boat as well, and most of the time I just make sure to take the new models for a test drive. Just trying to use them here and there as well, figuring out their capabilities and shortcomings. Makes it much easier to smell the vapourware when I hear the news.

The comments here are very focused on LLMs, which makes sense - that's where the hype is. If you really don't mind ignoring the nuts and bolts, you can treat all the large language models as black boxes that are getting incrementally better over time. They're not difficult to interact with from a developer perspective - you send text or tokens and you get back a text response.

It's definitely worth trying them out as a user just to see what they're capable (and incapable of). There are also some pretty interesting use cases for them for tasks that would be ridiculously complicated to develop from scratch and "it just works" (ignoring prompt poisoning). Think parsing and summarizing. If you're an app developer, look into edge models and what they can do.

Otherwise dip your toes in other model types - image classification and object recognition are also still getting better. Mobile image processing is driven by ML models at this point. This is my research domain and ResNet and UNet are still ubiquitous architectures.

If you want to be sceptical, ignore AI and read ML instead, and understand these algorithms are just another tool you can reach for. They're not "intelligent".

Hey Swyx, I'm a dev who did your (excellent!) email LLM course, so maybe I can give some info. I'm in the Latent Space Discord and have been trying to figure out what's next after the course. The challenge I've found is that most online discussions about LLMs are either very basic or assume a fair amount of context (true for the Latent Space discussion/podcast, as well as Karpathy's videos).

I've been trying to find the best next step and what seems fruitful from my vantage point are:

1. Cohere's LLM University - Seems to go more in depth into terms like embeddings that are still pretty unclear to me. 2. promptingguide.ai - For similar reasons, that it covers terms and concepts I see a lot but don't know much about. 3. Reading survey-level papers.

I'm including this info just in case it's useful to you, as I've really appreciated all the content you've put together.

One specific thing you or someone else could do that is simple yet high value is to create a list of "the first 20 LLM papers you should read". I've looked for this to build out more base knowledge, but have yet to find it. Suspect it would be helpful to others as well.

I spent some time trying to get chatgpt to write a front end in js. It would plot using a library and then when I complained about a bug it would say "Oh you're right, this library does not implement that commonly implemented method, instead use this code." and then would get in a circle of spitting out buggy code, fixing a bug, and then reintroducing an old bug.

It was okay, but kind of annoying. I understand js well enough to just debug the code myself, but I wanted it to spit out some boilerplate that worked. I can't remember if this was chatgpt omni, I was using or if it was still 3.5. It's been a short while.

Anyways, it is cool tech, but I don't feel like it offers the same predictive abilities as class ML involving fits, validation, model selection etc for very specific feature sets.

The question is what does programming with an LLM get you over batteries-included frameworks with scaffolding like Rails or Django? If the problem only requires a generic infra solution put together by an LLM instead of a bespoke setup, why not look into low-code/no-code PaaS solutions to start with? Unless the LLM is going to provide you with some uniquely better results than existing tools designed to solve the same problems, it feels like a waste of resources to employ GPUs to do what templates/convention-over-configuration/autocomplete/etc already did.

The point isn't that LLMs are useless, or that they aren't interesting technology in the abstract. The point is that aside from the very real entertainment value of being able to conjure artwork apparently out of thin air, when it comes to solving practical problems in the tech space, it's not clear that they are achieving significantly more - faster or cheaper - than existing tools and methods already did.

You're right that it's probably too early to have data to prove their utility either way, but given how much time, money and energy many companies have already sunk into this - precisely without any evidence to prove it's worthwhile - it does come across rather more like a hype cycle at the moment.