I'm old enough to remember how it worked for the Internet. It was treated as a curiosity for nerds for a long while. The whole time I thought it was completely underrated and would change everything - thats why I decided to get into tech.

Then there was a short, sharp "it's going to change everything" hype cycle that lasted no more than a few years followed by a trough of disillusionment that coincided with the recession that was around 2003 or so.

I thought almost all parts of the cycle were exaggerated from the part where it was a curiosity for nerds to the "pets.com craze" where "the high street would disappear completely".

With LLMs it's very similar but rather than a meaningful shift in technological development underpinning it it's something more like a parlor trick with a few practical applications.

Then sadly big business and government figured out how to turn that promise into the dystopia of today.

a parlor trick lol. Here we are seeing statistics applied on abstract terms as human knowledge, how this is something with few practical applications?

Because the results and actual efficiency gains are flaky, inconsistent and mostly limited to boilerplate?

GPT and friends are very impressive, yes; if you’re also impressed by your intern blindly copying stack overflow.

Maybe I am a fun of fuzzy logic, but I find gpt less flaky than most people that I read about, I think it will evolve. I feel gpt will even allow us to create new branch of mathematics applied at psychology.

I can't understand why you compare it with an intern copy paste, it's simply absurd to compare. Looks just a way of reducing the argument and drawing parallelism where there are none. Yes, some people use it to program, No, not everyone uses it for programming

> With LLMs it's very similar but rather than a meaningful shift in technological development underpinning it it's something more like a parlor trick with a few practical applications.

What makes you think this, I'm assuming you've tried gpt-4? I've personally gotten a lot of productivity out of it. Assuming it gets a lot better, I don't see how it's not going to minimally become an indispensable tool, and at maximum completely transform the future of humanity.

That would be a safe assumption, yes. Im clearly not a complete moron.

It reduced my productivity when I used it.

Between having a slightly faster and slightly customized version of google vs. being sent down misleading rabbit holes, the latter won out.

I found a few niche use cases where it proved useful but it was all out of proportion to the hype.

I've paired with other people who were bullish about LLMs using them and witnessed them falling down rabbit holes. I found the whole experience bizarre - it was like seeing a new "social reality" take precedence over what was happening in front of our eyes. Computer says yes.

LLMs are neural networks. If you understand their strengths and weaknesses they are extremely useful. Use LLM's to explain abstract concepts, discuss ideas, analyze text, as an interactive tutor, not as a datastore of facts to be recited verbatim (like a search engine). That is the worst possible way to use LLMs and will result in hallucinated facts. If that's how you're doing it then you're doing it wrong. Neural networks such as LLMs are fundamentally not made for factual recall. LLMs are designed for solving natural language understanding tasks.

I have found GPT-4 very useful for understanding concepts and solving specific problems in programming, science, maths, psychology, relationships, and producing creative writing, by having conversations with it, going back and forth deeper into these topics. But I would never use it as an API reference. Raise it up to the conceptual level and you will be surprised at what it is capable of.

That’s a ridiculous statement. Hydrogen fuel cells will eventually power all cars.