Hacker News Reader

You can criticize LLMs all you want, they will still be a huge thing in our economy and civilization.

Maybe so. That being said, it's been 3 years since Text to Transformer was released and... we still don't have any novel applications.

I'll take my chances.

Really? What about question answering in line in google searches, chat gpt, that chrome extension demo that can navigate web pages. Is the idea that these aren’t novel?

They're neat, but they're old. And frankly, they don't illustrate enough of an opportunity to build a product off of. Worst of all, they're not nearly dependable enough to provide any guarantee for. At best, you're building on a lie generator, which is only useful insofar as fooling the user is concerned.

I honestly think the last 3 years have been more exciting for compute hardware than AI software, which is really saying something in a post-Moore's-law world.

I think they're only old in that there were terrible version of all that tech in the distant past, but now chat gpt is actually practically useful to me fairly often, machine translation is a bit older but has gotten much better. It's a bit like you looked at google search and went, well yahoo already exists this isn't actually new.

I guess we'll just have to see

Reply on news.ycombinator.com