[antonvs]

It’s one of the most amazing technological achievements of the 21st century to date, if not the most amazing by far. But it’s still a machine executing a program. Your emotional reaction to it might be because you’re seeing it as something more than that.

[runarberg]

mRNA vaccines takes the first place for sure, and by a wide margin. But also advancements in lithium ion batteries and solar cell technology. We also finally generated electricity with fusion power this century, traveled to Pluto and set up the James Webb Space telescope [albeit the latter two are engineering achievements].

I think using an absurd amount of power to train a neural network on stolen data which is then used to perform mundane tasks and performs significantly worse then if done manually. If anything LLMs is one of the least impressive, yet most annoying technological achievement this century.

[antonvs]

mRNA vaccines were developed in the 1990s, i.e. last century. None of the other things you mention come close, but I'll detail that at the end.

First, I want to point out how inconsistent you're being in mentioning "electricity with fusion power". We've had the ability to create fusion for almost 3/4 of a century. Most of the nuclear weapons in the world today are primarily fusion weapons (mostly with fission starters.) This is a technology that has killed nearly a quarter million people. But you're touting a variation of it as being one of the most amazing technological achievements this century.

That's fine, but then you have to recognize that you're distinguishing between the capabilities of the technology and its actual usages.

Now do the same thing with AI, and if you're intellectually honest about it, you'll be forced to change your position. The technology represents the greatest technological advance of anything we've seen this century. The political and economic forces surrounding its use are a separate question.

As for your list of technologies - the problem with it is that none of those items seemed like science fiction at the start of this century. Whereas, what LLMs can do now was essentially science fiction as little as five years ago.

In detail:

* mRNA vaccines were developed in the 1990s. There was little question that they could be made to work in humans, all that was needed was sufficient impetus, which was provided by COVID-19.

* Advancements in lithium ion batteries and solar cell technology are incremental enhancements to existing tech. No comparison here.

* Generated electricity with fusion power - that was first done in 1991 at the Joint European Torus (JET) facility. You may be thinking of the so-called "scientific breakeven" which was achieved in late 2022. That didn't actually generate electricity, just heat. It was also a rather arbitrary milestone that doesn't actually mean much, in practice. We're still orders of magnitude away from a true total system breakeven, where the total energy produced is greater than the energy consumed. The system that achieved scientific breakeven is not the kind of system that can be scaled up to address that gap. We're even further away from commercial power breakeven, where the value of the energy produced is greater than the cost of producing it. It's quite possible, even highly likely, that neither of the two goals I've just described will ever be achieved. There's no real breakthrough here, just marketing hype in a multi-decade effort to achieve something that may not even be possible.

* Traveled to Pluto - the Voyager spacecraft passed Pluto's average orbital distance in the early 1990s. Choosing to send a probe directly to Pluto is just a choice that would have been just as possible last century.

* The James Webb Space telescope - certainly a nice incremental advance, but not in the ballpark in terms of achieving something we didn't even know how to do before.

In short, if that's the best list you can come up with, it essentially confirms that LLMs are the greatest technological achievement of the 21st century so far.

[runarberg]

You are being overly obtuse. LLMs is also a technology of the 1990s with many minor breakthroughs this century and finally a major breakthrough this decade. And these breakthroughs were all about finding novel algorithms to apply to the model. Neural networks and language models are ancient technology really.

The exact same story is with mRNA vaccines. And almost the same with fusion power, though it is still on the minor improvements phase, the latest being the first time we reached an energy gain factor in the National Ignition Facility experiment in California in 2022.

You are applying arbitrary limitations in what counts as “new technology” and then selectively applying them such that LLMs have a lower barrier then other technology.