It was never accurate. The concatenated strings just failed to conform to reality as it often does. People are projecting too hard on a chatbot. It's just a chatbot people.
And I never seen lower entry barriers improve quality of systems. Also, I don't think generating code in a probabilistic manner is a win.
I'm not in a camp. What is it with current discourse that only sees adversary teams as valid positions?
Great counter-point. It is a false dichotomy to view this as a binary matter (terrible pun not intended).
During the information superhighway era of the nascent internet, opinions on what would become varied as well (I recall with a bit less sarcasm and absolutist thinking).[1]
The hype was not widespread, and even though some prognostications were fairly accurate in broadly outlining societal and economic benefits, predicting the downsides was an oversight.
Maybe failing to foresee these difficulties in the past has left some people jaded? Could this be driving much of the cynicism and skepticism in the great AI debate?
It's already useful to me as my job involves writing reports to various audiences.
Taking to a computer(programming) is like using a knife to undo screws. It works mostly but you need to be careful.
I think you have to separate what's hype from what's not. I personally see three things here:
- The usual hype, now AI, previously Bitcoin/NFT, some time earlier it was "sharing economy"[0] / "gig economy", earlier it was the cloud. Some things change, some stay the same, but there's little point paying attention to it unless you're in for a quick buck - for the shrewd, that hype is effectively a pyramid scheme with extra steps.
- The fear of losing your job to the AI. This, IMHO, is a quite realistic fear - more so than ever - because GPT-3.5 is already[1] good enough to replace plenty of jobs. It's just the market hasn't caught up to it yet. But it will soon enough.
- GAI, consciousness. There's no hype there. It may seem there is, because you're part of the small group that cares and can productively discuss the topic; for most people, it flies entirely over their heads. On that topic, however, I feel quite strongly[2] that LLMs are a major breakthrough in understanding the nature of intelligence. Disregarding the language model aspects, whatever is going on in the latent space, it seems unexpectedly powerful.
My speculative take: it might be that intelligence is mostly a natural consequence of building a high-dimensional association map between sensory inputs. It would definitely make sense in the context of evolution, and the spectrum of intelligence that various forms of life fall on. Perhaps we didn't realize it before because we can't imagine high-dimensional spaces all that well.
Even if my hypothesis turns out to be false, I can't imagine something akin to LLMs, with high-dimensional latent space, not becoming one of the fundamental concepts in AI and cognitive science. I expect we'll find examples that fit this model in nature. I expect we'll find that a part of our brains fits that model too.
But back to the point, the GAI/cognitive/consciousness aspects are hardly hyped. They're just becoming hot again among the small crowd that is interested in this. As you say, it's usually that hype follows real change. I imagine if LLMs end up shedding light on the topic of consciousness, both hype and hate will follow, particularly from the world's religions.
--
[0] - $deity knows how crazy it was to hear the excitement about "shareconomy" coming from the startup/entrepreneurial crowd, seemingly oblivious to the fact that anything they do in this space will be abusing, exploiting and otherwise ruining it...
[1] - Consider all the things people use ChatGPT for. All the things that you may be using it for. There are plenty of jobs that are entirely contained within the "area of effectiveness" of ChatGPT. And I'm talking the GPT-3.5 ChatGPT here (sadly, I still don't have the API keys to GPT-4 :( ).
[2] - I'll give it 90% confidence, though I'm not exactly well calibrated.
What if consciousness actually is the high dimensional entity that can map to sensory input and motor output of neural networks? Then expecting it to emerge from highly manipulated memory spaces would be like shocking cadaver brains with lighting to get a person... Physics is looking more and more like it cannot actually describe an objective reality that is perceived, pointing more to reality being the descriptions that we can build of experience in high dimensional spaces.
But speculation apart, it doesn't much matter how we think of chatbots or what they can do. Ultimately all AI/ML stuff is just up front compute stored in models, and they are just forms of capital. They will be used as other forms of capital.
To think that we can build people surrogates because chatbots are getting impressive is... sales hype pure and simple. Much like thinking you don't need servers because you use "the cloud". You're just outsourcing engineering, something that has been done for centuries.
Maybe we should petition dang to allow memes to be posted and get it over with.
Seriously though, "stochastic parrot" is not a 4chan type of meme, it's the "industry term of art" kind of meme.
it's been used 6 times on /g/ and almost all are mocking the term -- with one calling people stochastic parrots (haha)
Meanwhile it's been cited 1200 or so times.
>> "Camp stochastic parrot"? HN is looking more and more like 4chan.
(https://news.ycombinator.com/item?id=35451667)
Then I clarified that:
>> I'm talking about your trying to split comments into camps.
(https://news.ycombinator.com/item?id=35451944)
Not that "stochastic parrots" is a 4chan thing. The polarised debate with "us" vs. "them" is what reminds me of 4chan. And not just 4chan, it's how people generally behave on the internet: it's always "us" vs "them" and there's no space for interesting conversation.
Extreme Polarization in debate FAR easily predates imageboards and even USENET, vi vs emacs probably harkens back to some of the earliest messages on DARPANET (ok, probably the text editor wars were close to a decade later, because i think it was more like the "ed" vs "god do we have anything else" era, but you get the idea -- i'm sadly not old enough to know the details of this golden era)
or look upon ye olde days of Slashdot trolling (some netcraftian oracles say that BSD is still, indeed, dying)
if you really do want to track the modern internet manifestation back you have to go back before 4chan. We'd have to talk about SA and GBS vs FYAD, and some of the earlier predecessors of SA such as.... Maddox.. and others, i guess? I can't think of other sites, only other names.
In any case - don't give 4chan credit where it's undue. Us vs Them goes back a long, long while, perhaps even pre-internet!
Anyway, what's with the assumptions? "Lad"? A lad named YeGoblynQueenne? Not that there isn't precedent for that sort of thing [1], but, let's not jump to conclusions.
______________
Allegedly, HN is a place to hold curious conversation. What is curious in attempting to explain a comment as coming from a certain "camp"? Isn't that just going to polarise conversations and shut down any curious discussion?
> You can be in many camps at once, there is more than two camps and they don't have to be necessarily adversarial towards each other either. However, if you think that LLMs are just stochastic parrots then that probably guides a lot of other arguments and lines of thinking.
We are a long ways off from a flame war and 4chan here.
Making this sort of assumption about your interlocutor deprives the discussion of any kind of interest it would have. It's like we have seen that match before and we know what each side is going to say, so what's the point in debating anything? Everyone's convinced, everyone's stuck in their little corner, and calling each other group-names.
More to the point, it makes people reply not to each other, but to the views they assume the other person holds, which never leads to productive conversation, only to confusion and chaos.
There has to be a way to debate that doesn't involve picking an enemy to fight.
it's been used 6 times on /g/ -- 3/6 are mocking the term, 1/6 is calling people stochastic parrots (haha), and 2/6 are using it in the sense you're implying.
So not really a 4chan type thing. It's been cited by other papers apparently 200x more than it's been posted there.