[cpeth]

In terms of impact, LLMs might be the biggest leap forward in computing history, surpassing the internet and mobile computing. And we are just at the dawn of it. Even if not full AGI, computers can now understand humans and reason. The excitement is justified.

[MrVandemar]

Nah. LLM's are hype-machines capable of writing their own hype.

Q: What's the difference between a car salesman and an LLM?

A: The car salesman knows they're lying to you.

[MVissers]

Who says the LLM’s don’t know?

Testing with GPT-4 showed that they were clearly capable of knowingly lying.

[nativeit]

This is all devolving into layers of semantics, but, “…capable of knowingly lying,” is not the same as “knows when it’s lying,” and I think the latter is far more problematic.

[SamPatt]

Nonsense. I was a semi-technical writer who went from only making static websites to building fully interactive Javascript apps in a few weeks when I first got ChatGPT. I enjoyed it so much I'm now switching careers into software development.

GPT-4 is the best tutor and troubleshooter I've ever had. If it's not useful to you then I'm guessing you're either using it wrong or you're never trying anything new / challenging.

[dilDDoS]

> If it's not useful to you then I'm guessing you're either using it wrong or you're never trying anything new / challenging.

That’s a bold statement coming from someone with (respectfully) not very much experience with programming. I’ve tried using GPT-4 for my work that involves firmware engineering, as well as some design questions regarding backend web services in Go, and it was pretty unhelpful in both cases (and at times dangerous in memory constrained environments). That being said, I’m not willing to write it off completely. I’m sure it’s useful for some like yourself and not useful for others like me. But ultimately the world of programming extends way beyond JavaScript apps. Especially when it comes to things that are new and challenging.

[SamPatt]

I don't mean new and challenging in some general sense, I mean new and challenging to you personally.

I have no doubt someone with more experience such as yourself will find GPT-4 less useful for your highly specialized work.

The next time you are a beginner again - not necessarily even in technical work - give it a try.

[seabird]

Smoothing over the first few hundred hours of the process but doing increasingly little over the next 20,000 is hardly revolutionary. LLMs are a useful documentation interface, but struggle to take even simple problems to the hole, let alone do something truly novel. There's no reason to believe they'll necessarily lead to AGI. This stuff may seem earth-shattering to the layman or paper pusher, but it doesn't even begin to scratch the surface of what even I (who I would consider to be of little talent or prowess) can do. It mostly just gums up the front page of HN.

[SamPatt]

>Smoothing over the first few hundred hours of the process but doing increasingly little over the next 20,000 is hardly revolutionary.

I disagree with this characterization, but even if it were true I believe it's still revolutionary.

A mentor that can competently get anyone hundreds of hours of individualized instruction in any new field is nearly priceless.

Do you remember what it feels like to try something completely new and challenging? Many people never even try because it's so daunting. Now you've got a coach that can talk you through it every step of the way, and is incredible at troubleshooting.

[MrVandemar]

>If it's not useful to you then I'm guessing you're either using it wrong or you're never trying anything new / challenging.

Please quote me where I say it wasn't useful, and respond directly.

Please quote me where I say I had problems using it, or give any indications I was using it wrong, and respond directly.

Please quote me where I state a conservative attitude towards anything new or challenging, and respond directly.

Except I never did or said any of those things. Are you "hallucinating"?

[cthalupa]

'Understand' and 'reason' are pretty loaded terms.

I think many people would disagree with you that LLMs can truly do either.