> on-premise GPU cluster

Renting a GPU server from a cloud and hosting your own llama.cpp is the path of least resistance.

> in the American business model

AI company valuations won't survive if they're only for the "American business model".

That's exactly what RLHF is for.

(In fact, "that colleague" might have even been the source of the RLHF training set.)

Driverless airplanes and driverless trains existed for many, many decades.

There's still a highly paid and trained guy in the cabin though, for obvious reasons.

Cars aren't any different. Do you really care about the exact kind of autopilot software when flying?

Someone should document these cringe comments for posterity. As a warning to future generations.

It's too bad the previous singularity cult manias 10 and 15 years ago weren't documented and are lost in time.

Ten years ago, most of the skeptics in these threads would have said the success of AlphaFold 3 was impossible.

It doesn't mean everything you can think of is possible.

People who say every jobs will be automated never held any tool more advanced than a screwdriver and never worked a single day outside of an air conditioned office.

They were predicting flying cars for "the year 2000"...

But enough jobs are going to be automated that it will cause social and economic chaos, and political breakdown. Sure AI can't attach an air conditioning unit outside my window nor use a plunger on my toilet but it can still replace lawyers, secretaries, software devs, and many more jobs.

I'm curious: how will it replace e.g. lawyers? Will judges be replaced as well? Juries?

Are we sur eof any of that?

Taxis were supposed to be all replaced when Uber promised to buy every single Tesla model S coming out of factories from 2020,they're not even self driving today...

Software engineers were supposed to disappear "in 6 months" every 6 months since chat gpt was released

Lots of people on this website live in insane bubbles which are completely out of sync with the reality of most workers

They'll only get to claim a hollow victory because AGI is impossible to rigorously define. None of the regulating bodies of consumer products will be able to define it better than academics. They'll use marketing to make those claims and there will be legal battles that keep it in a gray area.

They'll go for that because it's easier than actually inventing the 'sci-fi' AGI, shareholders keep making money, and it keeps them getting paid to keep going. If any of them actually do succeed, then that little deception will be peanuts.

Demis' bar is high and he stated clearly multiple times: AGI should be capable of inventing truly novel things. Examples he gave included the Theory of General Relativity and the game of Go.

Just a taste of what's to come: https://openai.com/index/model-disproves-discrete-geometry-c...

Anthropic's latest model also solved this 80-year-old problem that eluded many expert mathematicians, in a different way, according to one of its employees.

Remember we still have 4 years until 2030.

Surely you can think of some times the skeptics have been right.

I immediately thought of this:

https://en.wikipedia.org/wiki/List_of_predictions_for_autono...

Ten years ago, most people would have said cold fusion was impossible... and still is.

You know in the future they are just going to say it's different this time right?

If you look how much goalpost has been moved over the last decade it'd turn out that AI cultist were right more frequently than AI sceptics.

I'm sure HN cult comments from 10 and 15 years ago are still up though, the Dropbox thread is frequently linked to.

Just search for "stochastic parrot" on Algolia. Plenty of "BuT iTs NoT rEaLlY iNtElIgEnT" comments...

It literally is a stochastic parrot.

P.S. Yes, I do know how an LLM works. I code inference engines at my day job. You're consuming a cloud chat bot.

The people saying "it's just a stochastic parrot" were saying that because they thought the fact that it works by modelling the statistics of words and predicting the next most likely one meant that it would never be capable of anything sophisticated or any kind of novel output. It was just parroting things that already existed in a random way.

There was never any actual foundation for that belief, and it has been proven wrong empirically by more capable models, which is why you don't hear people saying it much any more.

Yeah, same.

DNS isn't centralized at all.

The top-level domain registrars are centralized. But you don't need to use them - you're free to use your own TLD's instead of, or even in parallel to, the official ones.

The moronic "no need programmers" hype cycle happens every 15 years. We've all been here many times before.

> AI is better at this than you. You just won't admit it.

No, you're just really, really shit at programming. You just won't admit it.

(AI, in general, is only impressive when you have no clue about the subject domain.)

LLMs are always amazing doing stuff you're bad at. (And always underwhelming doing stuff you know.)