[linotype]

> generative AI is a product with no mass-market utility - at least on the scale of truly revolutionary movements

This line is absurd. I use it constantly. 4o reads my code and generates documentation and type annotations. It generates boilerplate code. It generates logos for projects. I review all of its outputs code wise and make the odd correction here or there. I use it to check over documents before I send them. It’s replaced Stack Overflow entirely in my workflow.

I’m curious as to what’s above the author’s line for revolutionary.

[plasticeagle]

The fascinating thing with arguments like this is how they entirely ignore that OpenAI trained the tool you're using with Stack Overflow. And so if what you're suggesting is that AI tools will replace SO, then as soon as it does the AI is dead in the water.

Who will continue to feed it new information? The answer is, nobody will. And instead of having a community driven knowledge base, you'll have a bankrupt corporation and nowhere to turn to when you discover that you can't write code anymore.

None of the tools can exist without stealing the net sum of human knowledge - insofar as that is represented by the contents of the internet - for corporate profit. If what the AI proponents claim comes true, that source of knowledge will cease to exist. And what then?

[jononor]

The AI does not loose its utility once it has replaced SO. Further improvements would need to come from other sources than more "free" text data though. There are several avenues for this: A) better learning efficiency - it is known that models are not samples efficient now B) pay people to create more data at large scale - already happening at the big LLM companies (though maybe less do for code right now) C) learn from the prompting interactions that customers - extremely desirable, as people would effectively pay you to improve your model. Companies are gonna keep that close to their chest D) multi-modal learning and transfer between modalities E) Self-learning, where ML model iteratively improved itself. Analogous to AlphaGo, GANs etc. This is the holy grail, but also possibly the hardest. Code has the benefit of being formal languages with available verifiers, which should be useful here. F) Specific go code: Using functional feedback, not just textual. For example the large amount of unit tests available on GitHub

So I believe there are many avenues for further improvement. I still think it will be very hard though, and until we hit the next breakthrough, will be very resource intensive and possibly quite slow going.

[linotype]

> And instead of having a community driven knowledge base, you'll have a bankrupt corporation and nowhere to turn to when you discover that you can't write code anymore.

I still write a lot of code without needing to use an LLM. It’s just a tool that helps me be more productive. I don’t see how what is doing is any different than Google just displaying the answer instead of links to the answer.