[pdimitar]

Over the course of the last 3 years I found myself clicking on less and less HN stories. This place definitely feels like it lost the inventor spirit and the clever "how to" articles. A lot of stuff is either political or has very close political outcomes, and the rest is blind hype (ChatGPT won't revolutionize almost anything, and I'll stand by my words, bookmark this comment and show it to me in 5 years, I dare you). Let's not forget the yet-another-500-comments thread arguing pointlessly about whether startups are a good format or not. Or office vs. remote. And several others (but they are not many and it does feel like a tool a la ChatGPT could have generated them).

Today I've gotten much bigger value for my time when reading about various CLI tools that process and ingest / export data (recent thread about `miller`) than all of the above, combined.

HN, I feel, became more popular, and that has hurt its quality. So yes, I started reducing my consumption of it as well. I treat it like all other news sites 99% of the time, and I am right to do so at least 90% of them. I barely find 2-3 good articles per week these days.

Soon I might start checking HN biweekly because the value proposition is just not there.

[robswc]

>ChatGPT won't revolutionize almost anything, and I'll stand by my words, bookmark this comment and show it to me in 5 years, I dare you

1. I'll actually take you up on this, because I'm interested to re-visit the discourse around launch, 5 years from now :)

2. I do agree it's overhyped to a degree. I think it will revolutionize _some_ things but it could turn out like VR. I do use it every day and I don't see a reason not to... I guess we'll see what the future holds!

[pdimitar]

My problem with the hyped up stuff, ChatGPT in particular, is that AI actually does not exist.

The practitioners of the area of course have a vested interest to argue until the end of days that ACTUALLY AI does equal ML/DL and stuff but I am like "I see no Skynet so get off my back, we have no AI and that's that". :D

But, I guess in 5 years we could argue whether certain progresses are indeed attributed to ChatGPT or is it something else entirely!

[CamperBob2]

If you don't know how your program works, in the sense that you don't know how it arrived at the answer it gave you, IMHO that's as good a reason as any to call it "AI."

The days of understanding our own code are rapidly coming to an end. As developers, we now have the same problem that the mathematical community has had to deal with for some time, as proofs become complex enough to demand large chunks of peoples' careers to comprehend and evaluate.

Prompts may or may not be the next revolution in the graphic arts, but they will be the next revolution in programming languages. I've seen more than enough to conclude that.

[pdimitar]

> The days of understanding our own code are rapidly coming to an end.

That's one possible development of events, yes. The other one is inventing DSLs on top of popular languages so we're spared tons of boilerplate; but then we'll need experts on the actual underlying languages (say C++, Rust, Lua, Erlang or anything with a good VM and/or statically strongly typed and compiled to native code) when sh1t hits the fan, which... actually can work just fine.

[netsroht]

This is why I prefer the term "weak AI". Weak AI is specifically trained to solve tasks whereas strong AI can teach itself to solve new tasks.

Whether humans are able to create strong AI is a philosophical question: While some argue that's not possible (can we be Gods?), others argue that this is the next logical evolutionary step.

Let's see if we can at least mimic strong AI when we let LLMs connect to external systems (internet, money, more energy, etc) and specifically allow themselves to fine-tune or train new NNs in general.

Time will tell.

[pdimitar]

Well comments like yours are the ones truly deserving discussion on the topic, just so you know my opinion. <3

> This is why I prefer the term "weak AI". Weak AI is specifically trained to solve tasks whereas strong AI can teach itself to solve new tasks.

Yes, we can call it a spectrum, though I'd think it's more like a multi-dimensional space. Wouldn't contest your definition though, it's as valid as all the others really.

And yeah I agree/think that the ultimate general AI is the one that can teach itself new tasks, utilize past experience even if the patterns don't match perfectly, and have some sort of sentience. And let's not forget that it must have actual goals and motivation (otherwise it'll just conclude that the best course of action is to not expend any effort and just put itself in an infinite idle loop).

> Whether humans are able to create strong AI is a philosophical question: While some argue that's not possible (can we be Gods?), others argue that this is the next logical evolutionary step.

IMO people romanticize these topics too much. The true AI will be "born" as a hyper-optimizing recursive machine (and collection of algorithms) and it will eventually get limited by the physical reality it inhabits so it'll self-balance quite fine. It's strange how much spiritual value people put into these things, though I somewhat understand; that'll be the second truly intelligent and sentient "life form" that we will know beside ourselves so some metaphysical hand-waving seems unavoidable and maybe even desirable (in terms of moral correction mechanism, maybe).

> Let's see if we can at least mimic strong AI when we let LLMs connect to external systems (internet, money, more energy, etc) and specifically allow themselves to fine-tune or train new NNs in general.

I have no doubt we'll eventually get there but my feeling is that the current "AI" area is in a local maxima and it won't crawl out of it easily.

[galaxyLogic]

It's very artificial you know. There's this connotation with the word "artificial" that it is not real, but artificial.

[username3]

Can you favorite the 2-3 good articles you find, so we can follow your recommendations?

https://news.ycombinator.com/favorites?id=pdimitar

[pdimitar]

Eh, favorites to me really means favorites, meaning I don't want too much in there.

[IIAOPSW]

Replying here so I have a url for the remindme bot. see you in 5 years.

[x49asvk]

See you in 5 years! Though I almost agree with ChatGPT not revolutionalizing anything.

[IIAOPSW]

Depends how loosely we interpret this. Do we mean online AI language models which are part of a continuous lineage tracing back to the present chatGPT will not have revolutionized anything, or do we mean that OpenAI and their product line known as chatGPT in particular will not have revolutionized anything? Let's disambiguate now so no one says we are equivocating in 5 years. I'm understanding it to mean the former.

[pdimitar]

I mean the latter. Science progresses with big steps every now and then so it would be crazy for me to claim "AI will never exist". It's clear that eventually we'll get there.

It's a fact that somebody somewhere thought of making better bearings for their horse-pulled carriages centuries ago. That person does not get credit for modern cars' suspension systems however.