reply on: There Are No New Ideas in AI… Only New Datasets \ stacker news ~AI

pull down to refresh

130 sats \ 2 replies \ @optimism 2 Jul \ on: There Are No New Ideas in AI… Only New Datasets AI

[T]he reality is actually cooler than [just figuring stuff out]: we’re engaged in a decentralized globalized exercise of Science, where findings are shared openly on ArXiv and at conferences and on social media and every month we’re getting incrementally smarter.

This is what pleases me about the AI scene: the research is not consolidated into a few companies or universities. I hope this can be retained, especially since the bulk of the debt and meager results are of course rather concentrated, and the overlords seem to be favoring mega-corporate interests (even when the senate votes 99% against state law prohibition.)

If you squint just a little, these four things (DNNs → Transformer LMs → RLHF → Reasoning) summarize everything that’s happened in AI.

Squinting a lot there. There are many things that just aren't realized yet by the corpo dudes but that are out there as research and PoC code that I use every day. The scene is not just at the base models; it's very much about the application of the models. We cannot (and should not, imho) rely on OpenAI to dictate the product roadmap for the entire technology.

There are millions of retards like me that just yolo some code together and get something running and maybe it works. It doesn't matter that this isn't some AI-as-a-Service offering. It could become that but then we get back to Zuck-style abuse of the masses. So let's just slug on.

100 sats \ 1 reply \ @SimpleStacker 2 Jul

There are millions of retards like me that just yolo some code together and get something running and maybe it works

I feel like 99% of companies calling themselves "AI powered" is basically this. They just threw together a ChatGPT wrapper and call themselves AI-powered.

This is what pleases me about the AI scene: the research is not consolidated into a few companies or universities

Is this really true? The capital costs to get started on AI research seem immense

30 sats \ 0 replies \ @optimism 2 Jul

I feel like 99% of companies calling themselves "AI powered" is basically this.

Worse, I was talking to a friend yesterday, their AI-powered technology provider just uses Claude for everything and now they sell themselves as super smart. They're just vibe coders but the general public does not know the difference until the contracts are signed.

Is this really true? The capital costs to get started on AI research seem immense

Depends on what you research: if you're looking to train a fully fledged model to replace OpenAI's stuff, sure. But what if you're doing research around peripherals? E.g. #1017613 could potentially be researched on top of an existing open distilled model - something you could in theory run on a modern 4M sats mini pc as long as it has a proper GPU.