GPT4 simulating a FTP server at ftp.disney.com
filestash.app
filestash.app
Hallucinating it would be if it came up with the concept all by itself. Like I asked where I could see the latest Avatar film and it said "easy, just go onto ftp.disney.com and download it".
The year is 2026
"hey, how can I watch 'avatar 3 - tokyo drift' today? it's not out yet." "easy, just go onto ftp.disney.com and download it". "really?? can you do it for me?" "yeah here it is"
suspiciously fast response arrives, with a video which is a movie about blue aliens that gpt6 has just asked midjourney 8 to create for it.
In a not so distant future you will create your own movies.
Or even better: There will be models teached on a show you really liked but which was cancelled and you will be able to continue it to eternity.
You can't take the sky from me~
"Firefly season 2 by rustyBones447 is the best one"
Not really challenges, but business models. There is no free lunch so to say. Maybe there will be highly personalized subscriptions that allow for hyperparameter tuning in this way.
I doubt this is something Netflix fears, it is something they gonna embrace. Computation of such movies won't be cheap.
A large part of value of movies, TV shows, books, videogames - even paintings and sculptures - is that they're part of external and shared reality. With self-generated movies (shows, books, games, ...), there will be no process of discovery to enjoy, no immersion to experience. There will be nothing to share or discuss with others, as everyone will have their own version of the story. There will be no canonical version to talk about - there will only be everyone's headcanon, all equivalent, all personal. All boring and worthless.
Now imagine if anyone could casually just fork a piece of writing and adapt to their personal tastes. "Like this, but make X be in love with Y instead of Z. Also make Z not really die but make a surprise comeback in the final chapter." It might be fun to play with such capability for a moment or two, but I can't imagine anyone enjoying exchanging such stories - if everyone can trivially change anything, then everything becomes arbitrary. There is no story, only everyone looking at their own selfie.
Incidentally, GPT-3.5 and GPT-4 already give us this capability with smaller pieces of fiction. You can literally alter any short story like I described above, today. Anyone can. And somehow, I don't see any takers. I think people realize there's no entertainment value in it. Hence, this is what might just save Hollywood et al.
Erotica is one exception to what I described above, as it's by nature the kind of fiction to be enjoyed personally, not in a wider context. And, of course, erotic story generators are as old as the Internet.
Writing, even prompting, is more work than reading, and it takes some of the fun out of it - plot twists are no fun if you told someone what they should be.
Plain ChatGPT still requires a bit back and force to get complex tasks going, but Auto-GPT already has a loop around it that makes it possible for GPT to feed into itself and perform complex tasks that are beyond a singular prompt line.
Even without that, ChatGPT can already provide quite serviceable short stories, with no effort on the user beyond "write a short story". The biggest practical show stopper right now is that ChatGPT has strict filtering, so the moment you have guns, action and death, the story just ends. BingChat is even more extreme with the filtering and will fail almost every time.
So no John Wick fanfic then
I don't think so. I think users will share what they've generated like they do now with midjourney images.
If I decided to "generate" a pilot episode of a show about robot ninjas fighting a secret war on mars, and it turns out really well, I'll share it on youtube. if enough people like it I might do ep2, or ep3. Then it becomes that shared reality.
Obviously if anyone can just say "gimme a show about robots on mars" instead of looking at what others have already created, the audience might not be as big, but no more than midjourney I'd imagine there will still be some talent and finesse involved in getting the prompts just right, and most people will be too lazy and prefer to browse what others have already done
Like it or not, on-demand fanfiction is where we're headed. It's not impossible that a canonical version of a story might be published though, like a video game. There's a main story about you saving the world from nuclear disaster, but if you spend 600 hours trying to romance the barfly in the first tavern and never fulfill your role, well, that's your story about a lecherous loser in the land of the lounge lizards, man.
This isn't anything new either; look at the Protestant Reformation. Everybody wanted a version of The Story that appealed to their own interests then too. Now you can pick a church whose ending you most prefer.
Would be awful.
The best movies I have seen I would not have been able to come up with.
https://i.ibb.co/XLFGdWp/DALL-E-2023-04-05-12-55-36-Mickey-M...
Mickey Mouse according to OpenAI is just a symptom, but the root of the issue is very deep if the people preparing the dataset intentionally decide to exclude examples due to commercial or political pressure.
https://preview.redd.it/n6wdyhyz9dra1.png?width=960&crop=sma...
https://preview.redd.it/9djj6tdflu3a1.png?width=1080&crop=sm...
https://www.reddit.com/r/midjourney/search/?q=mickey%20mouse...
And if we've learned anything about AI in the last 6 months we've learned:
1) AI is probably trained on data obtained without the creators permission
2) AI responses are very hard to control, and easily jailbroken
[1] https://petapixel.com/2023/04/03/midjourney-bans-ai-images-o...
He might just want to be able to visit china in the future
Wonder how much other open source software I am unaware of?
> that gpt6 has just asked midjourney 8
at the current rate gpt16 and midjourney --v 20 are more likely
I'll pay, just gimme a file to download
The truth is, we have no need for the notion as it applies to AI. It's just a catchy term that sounds great in the media. Nothing the algorithm spits out is a hallucination (or else ALL of it is) because, quite literally, It's all the same uncontrollable output.
It'd be as weird as saying your autocorrect feature "hallucinated" any time it presented the "wrong" word to you. (As if it's of clear mind the other times?!)
We speak of "hallucination" when GPT provides completions that are not connected to any facts it was "provided with" in either training data or prompts, but ultimately factuality makes no difference to GPT.
It ALWAYS just "simulates" text, that's what it does.
If you'd want to use these words to describe factual accuracy, or to express if GPT "performs" an action, as opposed to "faking" a result, you have to bear in mind either of these qualifications are nothing but a human interpretation of the completion.
It being "correct" more often than not with the content of it's simulation is just a logical, but ultimately accidental consequence of being so well fitted to simulate the structure of data it trained itself on.
And a "halucination" is nothing but this simulation deviating from the context you provided, that fundamentally was simulated by the same rules that produced any other content which didn't deviate.
A true hallucination would be, if I were to ask where can I get Disney songs, and it referred me to a non-existent FTP server.
It's entirely inaccurate...
It’s literally doing what it was asked to do.
Is "this word fits here" now synonymous to "fully aware"?
And like someone said before, perhaps we can think of the GPT models not so much as “artificial intelligence”, but more like a different kind of intelligence from what humans have. With strengths and weaknesses of its own.
You even ignored the quotes that OP put around ‘aware’.
As someone working with GPT-3 day-to-day at $dayjob, these words - usually reserved for sentient beings - are simply put typically the most apt to use in pretty much all contexts other than some sections of an academic paper. The fact that you yourself couldn’t get the sentence to work right is a testament to that.
Again, as someone that’s working with and having conversations about this day-to-day, we all find ourselves using these words, first kind of awkwardly, with laughs and air quotes, and we try to mentally reach for the best word to use. Eventually we all just know what we mean, and that’s that.
So, if I can ask directly, how do I talk about this online in a way that allows for productive conversations about the very obviously disruptive abilities of LLMs but that still avoids your snark?
The way it feels to me is that I have a fully formed concept almost immediately upon hearing or reading something then I employ language to try and convey the same concept into to someone else’s brain.
During that process additional ideas may occur to me, and that may augment the final output.
I experience some of what you describe, too, when dealing with ideas I already am familiar with, or explaining them to others. But when I'm trying to work through some novel thoughts, or am engaging in a conversation - including the one that has me writing this comment - I function in the way that's peculiarly similar to LLMs. That is, what pops into my head are words, phrases, sometimes entire sentences (often movie or videogame quotes). I pick one that fits best as an extension of what was already said - the previous words, the previous sentences, the response from the other party - and then repeat with the next completion, and the next, etc., until I feel it's time to stop.
I've joked for many years that, based on my personal experience and introspection, human brains at the highest level may just be glorified Markov chains. I sometimes still say that about LLMs - but I do so more as a compliment now. GPT-3 and GPT-4 may be just fancy autocomplete, but so are we. I feel we've brute-forced our way into a turning point in our understanding of what intelligence is and how it works.
Like I read your comment. I had thoughts during that process, but I read until the end, then I made a decision about what I thought, and that decision contained all the information that I want to convey in my response.
Now: when I am constructing my response, I am of course modulating which words might come next that would best fit into my language model to convey that concept, but the concept is an instantaneous understanding.
That instantaneous understanding may change during the process of articulation, but not fundamentally -- I think that any changes during the process of articulation could be considered "elaboration" rather than a reconceptualisation.
All the language that I just typed was designed to convey a concept that was more or less instantaneously formed and fully complete prior to my typing anything. The way in which I construct the language to articulate that concept may be very similar to the way in which a LLM constructs its responses, but I think the core difference is that original and instantaneous conceptual understanding of what it is trying to articulate.
Only moments before I write the word, do I know what that word will be, but I know the complete thought I want to convey. I don’t know if this is important, but my thoughts are not words, so I’m basically just describing them using words.
The difference of course is that the concepts can change and merge with each other, but what I'm referring to as an "instantaneous concept" that precedes linguistic expression is like an instantaneous graphical image of an elephant, that one must then put into words.
I think that a LLM doesn't have that. I remember Kasparov had some criticism of Deep Blue that it wasn't really "playing chess" (although I can't find it now). In the same way I think that LLMs aren't really conveying a concept, they're only doing the part that assembles the language. The concepts are all pre-existing in the same way that all the chess move possibilities pre-exist in the programming of a chess computer.
EDIT: In fact even the way I described concepts as "pre-existing" there might be giving it too much credit. Like the concept only arises on the receiving end, from a pool of concepts that were pre-existing but about which the LLM has no "awareness" whatsoever.
The question is, do you know its attributes before you describe them to yourself? To me, at least, an "elephant" feels like a pointer, a handle, a promise - I know it's a thing, but to know more than that, I have to dereference the pointer. And that usually means something that's getting close to stringing words together.
Possible confounder: I suffer from aphantasia[0]. I have nearly zero ability to perceive images in my mind, except when dreaming. Maybe it's easier to "know immediately what it [the elephant] is and all it's attributes", if it just pops into your head as a complete image. I, unfortunately, don't experience that.
> The difference of course is that the concepts can change and merge with each other, but what I'm referring to as an "instantaneous concept" that precedes linguistic expression (...)
> I think that a LLM doesn't have that.
Crazy idea: what if a concept is... a partial set of coordinates in the latent space? Enough to describe a pretty narrow region.
> I remember Kasparov had some criticism of Deep Blue that it wasn't really "playing chess" (although I can't find it now). In the same way I think that LLMs aren't really conveying a concept, they're only doing the part that assembles the language.
My tentative guess is that language will turn out to not be the fundamental part of what makes LLMs so powerful/spectacular. I think the magic is the latent space itself - the high-dimensional constructs where all kinds of associations end up represented as spatial distance. Language is just a serialization format[1] for latent space associations. By being fed lots and lots of text in training, LLMs build up the latent space, piece by piece, until there's enough dimensions to it that it is no longer a simple mapping between words learned from the training data. I think this is how understanding and concepts look like.
Also, I think it takes so much input text for LLMs to get to their current performance, because LLMs learn like babies - they soak up everything they can, letting the inputs self-arrange by correlations. We learn to understand what an elephant is by hearing stories about it, seeing it alive, and/or being told facts about it in a structured way (i.e. school). LLMs learn what elephant is by soaking up so much text that (the token representing) the word "elephant" is, in the latent space, close to all the words people used to describe its attributes - which themselves are close to other words with whatever various meanings and aspects we relate to those attributes, etc.
EDIT:
And connecting the two parts of the comment: LLMs are AI's poster child of aphantasia. If they have concepts, how else could they work with them if not by... articulating them? And when they articulate a concept, what makes us so sure they're just parroting associations, and not truly having a concept?
--
[0] - https://en.wikipedia.org/wiki/Aphantasia.
[1] - Quite literally, as communicating via any language happens via ingesting or producing a series of some kind of tokens (letters, words, ideographs, etc.) over time.
I recognize that mode. I have it to. I had it now; this comment is aiming to transmit a thought that came to me fully formed while reading your response. That said, it's a bit suspect how it popped up as if my mind was doing an associative search. It's also interesting that, however fully formed the thought might have been, it's also small, as if a conceptual token, and I usually have to run further associative lookups to extend it before the comment (or train of thought) feels satisfactory to me.
In the previous paragraph, the last two sentences were not part of expressing the fully-formed thought described earlier. I'm not lying to you, it's just I didn't expect to write the second half of the previous paragraph (and this paragraph) when I was writing the first one; they came from conceptual autocomplete.
> That instantaneous understanding may change during the process of articulation, but not fundamentally
For me, it quite often does. I rely on it happening often. Example cases:
- On HN, I post maybe 1/3 of the comments I write. 2/3 of the time, the act of trying to express my thoughts makes me realize that a) I'm wrong, and/or b) it's a bad idea to continue (e.g. I'm just procrastinating, or posting something impolite, or not useful, etc.). Of the 1/3 that survive long enough to be posted, I still delete some - as sometimes the realization from the previous sentence hits me only a couple seconds after posting.
- When facing difficult problems, I do a lot of internal monologue. In it, exploring whether certain phrases feel right, and kind of sculpting the chain of thought, is a critical part of the process.
- Because I often feel I lack the working memory to keep enough thoughts in flight, I often "talk" with myself via a text file - externalizing my internal monologue. That process also involves sculpting sentences, paragraphs or bullet points until they feel right, so it's arguably still articulation. And the more I think about it, the more this looks to me like the kind of tricks you do to work around the token limit / window of GPT-3 / GPT-4...
EDIT:
Ultimately, I think we're both doing thinking in similar ways, perhaps with different proportions between modes, and different conception of it.
One relevant conceptualization I have that I remembered just now: in conversational situations, I often feel like my subconsciousness is a Markov chain / LLM autocomplete, spitting out phrases, and my consciousness is busy judging them, and choosing whether or not to pass them to the actuators (speech organs, or fingers on the keyboard).
The two "models" - the autocomplete and the censor - is something I can clearly see in my mind when I'm processing a problem in a calm manner, or participating in a slow-paced discussion. But I've noticed that the more excited I get (or intoxicated), the weaker the "censor" part is; in the limit (when I'm really excited, angry, or drunk), I feel as if the Markov chain bit is feeding phrases straight to my mouth/fingers, cutting out the middleman of consciousness.
(Of course, what I wrote is all "how the algorithm feels from inside", so it's not very objective or reliable...)
Or at least I used to, because I only recently realized that, over the last few years, I've mostly switched to thinking in English only. I'm now in a weird state: my native language (Polish) is still easier for me to use, but also using it feels wrong somehow. Often enough, when talking to people around me, I notice I'm translating my own thoughts from English.
It might be that this process of learning and internalizing a second language, to the point it eventually took over my first one, is what changed my experience of the thinking process itself.
Ultimately, I think the definite answer would require solving the halting problem. The approximate answer we should be able to get when we let models interact with some complex environment over time. That is, the model predicts the next token, which may cause some change of a shared state, and then the model receives a response.
This seems like an answer to the question, "why is the LLM so compelling?" And not "what makes something conscious/aware?" Which is an important question but feels like not quite the right one to ask and then come to this conclusion. And it bothers me if I try to accept an idea where I don't know what conditions I'd reject it (eg, I don't know what evidence would cause me to reject this hypothesis, I can't see a test I can perform to next reject it).
ChatGPT 'predicts the next token' so much better than other 'next token predictors' that using that as the comparison for a low effort dismissal is like saying "what's to stop a human tracking a criminal by scent, humans have noses and dogs have noses so they must be the same".
If we agree that Markov chains aren't aware, then surely we also agree this definition of what makes them aware is either incomplete or incorrect.
Predicting the next token turns out to be quite profound, I'm just not convinced, at this time, it's sufficient to be aware of conscious. If you have a reason I should believe it, I'm all ears.
If you want to test how good a chatbot is at predicting tokens, well, you just have to test it. You can throw a Winograd schema challenge at it, or let it take part in code competition or exams, let it solve puzzles and all that. Quick simple test will for example show you that ChatGPT isn't any good at calculating, as the answers will just be wrong.
As for awareness, we know what ChatGPT is aware of: Its 4000 token history and whatever information it learned from the training data. That's the grand total of its existence, it doesn't get any other input, it doesn't perceive time, it doesn't learn, that history is all there is to it. And we can call that awareness as ChatGPT is able to perform correct prediction based on that token history.
For consciousness there are still a lot of basic parts missing, such as the perception of time, senses to act with some external world, the ability to think for itself without getting explicitly prompted and all that.
I think you and I are generally in agreement about consciousness and awareness. I think consciousness is an interaction of an agent with it's environment, rather than an agent in isolation. I guess that's why I find this attitude that consciousness is a matter of feeding enough data into a static training process confusing. I think it's a kind of is-ought problem where I ask about the "ought" and people express the "is" and we end up talking past each other.
But I don't really have a good hypothesis to disprove this either, so I'll have to meditate on that. I think it may be testable but it's challenging to model.
I wouldn't buy it, but you can sell it to plenty of people.
Because GPT often tells me to use programming libraries that I can find no record of anywhere and generates URLs of references that I can also find no reference of ever existing.
Transcript:
Can I give you a document and ask for a summary?
Yes, you can give me a document and ask for a summary. However, please note that as an AI language model, my ability to summarize a document may depend on the length and complexity of the document, as well as the quality and clarity of the writing. Additionally, please be aware that I am an automated system and may not always capture the main points or nuances of a document as accurately as a human reader.
How do I give you the document?
You can provide me with the document by copying and pasting the text into our conversation. Alternatively, you can upload the document to a cloud storage service such as Google Drive or Dropbox, and then share the link with me so that I can access and read the document. Please note that I will only be able to access public documents or those for which you provide me with appropriate access.
How do I share a document on a google drive with you?
To share a document on Google Drive with me, please follow these steps:
1. Upload the document to your Google Drive account.
2. Right-click on the document and select "Share" from the dropdown menu.
3. In the "Share with people and groups" box, type "chatgpt@openai.com" to add me as a collaborator.
4, Choose the appropriate sharing settings (e.g., "Can edit," "Can comment," or "Can view") for the document.
5. Click "Send" to share the document with me.
I have uploaded a document for you to summarize https://drive.google.com/file/d/...(redacted)
I'm sorry, but as an AI language model, I do not have access to external sources or links. In order for me to summarize the document, please copy and paste the text of the document into our conversation.
GPT responds that there is an upload button because a lot of interfaces are described that way around the web -- there is not likely a hard reference for its answer in regard to its own interface.
I also had it suggest there was emailing capability in the playground interface, or it would say that it emailed me something I asked for. It would claim to have abilities to do a bunch of stuff like this that it couldn't actually.
Also recently discovered it hallucinates dates and times and has 100% confidence they're correct
I am slowly coming to the conclusion that everything it tells you which you dont already know, is highly likely to be wrong. Question is, how useful it actually is, given that failure rate.
It feels like a person that really likes to talk, and has no regard for truth. If you catch an acquaintance of yours deliberately making things up just to sound eloquent, you'd put em into a "dont trust with anything"-category pretty quickly.
It can also be wrong about the things you already know.
My rule of thumb is if the cost of generation + verification + estimated correction time is higher than cost of doing it your self - avoid AI.
So if you have stuff where you already know what needs to be done depending on how well it fits in the AI tool workflow, it might be a good choice. If not just don't bother.
Amusingly enough, this is the exact strategy you should use for outsourcing development too.
Confabulation adding data which was not provided. Example making up the plot of an episode.
Hallucination to claim stimuli which was not provided. Claiming my prompt included something it did not.
Delusions refusing to admit an error or mistake and the best thing to do is reset the history and explicitly exclude the topic. Examples accusing me of spreading harmful disinformation when I brought up Kirstie Alley's death.
Often it's mistaken and it's none of those, you can simply correct it or direct it better.
I've experienced both, and deal with both in different ways.
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?
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.
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.
Maybe we should petition dang to allow memes to be posted and get it over with.
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.
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.
______________
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.
I use ChatGPT daily, but always specify the information I want formatted. (In bullet points, a table or for example a news article.)
Of course, that’s only possible if the context content exists in the first place. If not, then it’s back to the old-fashioned technique of figuring things out for myself and writing about them.
Like converting a code sample from one language to another, or one format to another (e.g. raw css to object syntax), or complex replaces where the regex gets tricky- also great at writing simple docs.
Also use it for micro tutorials if it's on a topic older than a couple years old.
https://cloud.typingmind.com/share/dc9b6a87-cd00-43e9-a7ea-6...
The result I then could paste into a text editor, save as .ical, and import straight to my calendar.
How do I set up some backup infrastructure for my home server? How does Btrfs handle read errors? How do I cook X if I don't have ingredient Y? What's the correct syntax for this command? How to analyse and interpret these measurements from this vermicomposting experiment (helping my girlfriend with statistics)? How do noise cancelling headphones REALLY work? Is it okay to have my samba.conf set like this (it wasn't)?
Well that feeling doesn't exist with GPT-4. So far we've always been able to come up with something, together. If you don't like the first answer, you ask more questions. You can dig deeper. You can tell it what your guess is, what your intuition tells you about the problem, where exactly your uncertainty lies, e.g.:
You: I don't understand phenomenon X.
GPT: Oh that's easy, X is just [parrots wikipedia].
You: That's fine, but I don't understand how X differs from Y, they sound like the same phenomena.
GPT: I see why you'd think that, but Y and X actually differ in this detail called Foo that makes all the difference.
You: I still don't get it, compare Foo to something similar that I know from normal life.
GPT: Okay, so Foo is like an elephant who's too large to drive a car.
You: Ooh, I get it now. Me : I don't understand this publicly documented AWS service with plenty of OSS examples, can you suggest how I would solve some edge case X
GPT : Oh that's easy, invents bullshit that sounds exactly like what I need
Me : Googles the shit the GPT came up with and can't find any references
Me : Maybe there is something similar - spends more time searching
Me : GPT you're wrong - your solution doesn't exist
GPT : Oh sorry, here's the correct solution, comes up with more bulshit
Me : Googles the shit that GPT came up with and can't find any references
Me : Starts googling and solving the problem on my own
Me : Finds out X can't be done with AWS service
Or
Me : Review and come up with improvements to this code : CP code
GPT : Suggest bullshit improvements, gives factually wrong reasoning, makes code worse
Me : Points out everything wrong
GPT : Suggest other bullshit improvements
Me : Points out everything wrong
GPT : Returns basically identical code to input
Me : Propose a design to this technical problem
GPT : Implements the solution in the most roundabout way and suboptimal way
My only use case so far is stuff like "translate this model to openapi schema" because it was faster than setting up the tool, and similar tasks that are easy to verify and boring to type.
"Can Ronald Reagan be considered a liberal?
Ronald Reagan is generally considered a conservative politician. He was a member of the Republican Party and served as the 40th President of the United States from 1981 to 1989.
Reagan's political philosophy was based on a belief in limited government, individual freedom, and free markets. He advocated for a reduction in government regulation and taxes, increased defense spending, and a strong national defense. He was also a strong supporter of traditional social values, including opposition to abortion and same-sex marriage.
While Reagan did take some positions that could be considered more liberal, such as his support for amnesty for undocumented immigrants and his advocacy for the abolition of nuclear weapons, overall his policies and beliefs aligned more with conservative ideology. Therefore, he is not typically considered a liberal."
Now ask this to a human. You will get a polarized answer "No, Reagan is a conservative icon!!!!!!" lacking any nuance.
Depends on the human. Ask a political science professor and the result will probably be pretty similar, plus an earful on the evolution of the two major US political parties' positions since the 1980s and how that complicates these kinds of retrospective judgements.
Maybe a real honeypot could send the attacker commands to the bot for an assessment. I wonder how it would compare to the existing assessment systems.
I did something similar but had it pretend to be a Windows machine, and for a specific type of corporation. The problem, or maybe opportunity, here is to have this happen live (as a user types commands in.) Currently the response rate is way too slow for that to happen, but probably will be mitigated as time passes.
Isn’t most creativity mashing things together that no one before thought of putting together?
I have to say that, although I still disagree with him, I was unable to get GPT to produce midjourney prompts which I would consider creative. Even when I prompted GPT to take the perspective of a known artist, it produced an endless series of banal landscapes, cliche compositions, overplayed metaphors, and insipid imagery.
Midjourney only really accels when a human is involved with the prompting.
I had to concede to him at the end of the discussion that /today's/ AI are not creative.
I firmly believe though that the inclusion of all mankind's perspectives possesses the raw material to generate creative ideas. My understanding is that current AI lack the ability to "turn off" the vanilla oversaturation of averages that pervade when you are exposed to all stimuli. One day, perhaps soon, there will be AI able to combine ideas they are exposed to without trying to combine ALL of the ideas at once.
What these models do is more similar to pastiche, if we really want to compare it to some technique.
But the value of the pastiche depends on the intent and its perception by the viewer(s), because there's nothing inherently original in it.
So basically what models produce has no value, unless we are able to attach some to it. [1]
If we asked chat-gpt to analyze something it has produced, it will probably say that it is "similar to" or "in resemblance of" but it's unlikely that it would say "this is the work of genius, how original! lovely!"
[1] edit: you don't read "pizza maker creates fabulous art on AI" but <person who's already in the business> won a contest of <some art form> submitting something created by <AI of your choosing>. Why? Because they know how to market it and can rely on other people believing that's their creation. Nobody says it upfront "I will submit an AI generated work", because they know it won't be judged the same way. The pizza maker was probably trying it for fun or to make a new logo for the pizza place or simply has no instrument to assign it a value and convince other people that it is true (including using their professional card).
When it is wrong it may correct itself, it may double down on being wrong and often just make something up again.
This isn't true. You can ask it basic logic problems that it's never seen before, and it will apply the rules of logic to them. It can also identify correctly which rules of logic would make sense to apply in more complex situations, even when it doesn't get the answer right straight away. At doing this stuff GPT4 is better than GPT 3.5 which is better than previous GPTs. I fully expect that future models will be able to tackle more complex applications of logic to new domains successfully.
If you use only examples that weren't in its training set, you'll get to its limits quickly, but basic level first order logic is definitely within its ability.
For example, it can take code and add types to it. This involves a lot of reasoning ability. It can do this because it’s been trained on a vast amount of code. But it can’t yet fully transfer those reasoning abilities outside the narrow domain of code.
For instance, OpenAI found that GPT4 is much better at reasoning in some human languages than others. It is best at reasoning in English, but struggles reasoning in less resourced languages.
There is clearly some context-independent reasoning going on (i.e. generalization) otherwise the model would not be able to reason at all in languages that it hasn’t seen a particular problem in. But there also appears to be a large context-dependent factor.
As you said, when things are not in its training set it can struggle. If there is a plausible looking text for the question I asked it will give it to me, that's how it is designed. For example I asked it about Windows command line debugger - CDB. It gave me an example command line for it: cdb -c "your-app" -o "logfile". It is very wrong. -c requires an argument which are debugging commands to run on start, -o is to attach to all created attached processes. Real command line looks something like this: cdb -logo "logfile" "your-app" (and it still does not exactly behave as you would imagine having experience with Unix CLI). The problem ChatGPT has with CDB is probably, because it has much much bigger corpus on Unix-like command line tools and because the documentation for CDB is abysmal. From this ChatGPT session I would have more examples.
I'm not saying it is useless. It just is not designed to do that. It might improve or there might be an another algorithm needed on top or instead of what it does use.
For me it is like a kind of a step up from a search engine. It helps me to find something to start with. When I get to some details it is often wrong. I get a starting point from it and then find a proper source for the rest.
In fiction it is often the case that an AI can reason better than humans do, but it doesn't understand emotions. But we now in general that emotions and reading of emotions is simpler than general problem solving. A child picks up on parent's emotions without extensive training. Animals can sense them. The fear response is a basic instinct. I would imagine it should be easier to make a machine being able to almost perfectly recognize emotions than general reasoning or this big heuristic machine which is GPT. I guess it all goes to the training set available.
If you write some simple code to generate art procedurally with randomness and it creates really beautiful pieces sometimes, was your software being creative?
I've played around a bit with getting it to write stories etc., and they do often seem quite "creative", the problem is eventually they often end up making little sense or contain fairly obvious contradictions or non-sequiturs in a way I wouldn't expect to see in the output of a typical human author (and certainly not a skilled writer). Indeed I'm not sure ChatGPT has the ability to formulate any sort of longer "story arc" within which to frame the text it generates. But I also suspect it will gradually be able to develop that capacity with future refinements.
Ironically I think anyone who believes LLMs cannot be creative are themselves uncreative in their prompt crafting and use of the model, I've even seen this from senior developers and experienced writers. It's painful, like watching your mom try to find the downloads folder or enter terrible search engine keywords.
I tested GPT on its ability to handle word creation with suffixes. Some results are really interesting, but sometime it pretends a word is a sample of some suffix use when the word doesn’t encompass it. I’m rather confident that this kind of artifact can be overcome, I won’t slander the technology because it still has some obvious weakness. But I’m afraid that once overcome, the lake of snake will convince most people all the more that this impressive technological achievement is more than what it is actually.
- Oscar Wilde
Algorithms like GPT or Stable Diffusion explore that n-dimensional space starting from possibly more places than a single human has experienced. They will miss many of the starting points that any human experiences, because of limits of those algorithms' interfaces with the world. However they can still find interesting places. Maybe it's not creativity as we mean it, but they can show us interesting places that no artist has found before.
Obviously we can say things like "What about a car that ran on pencils" but real, useful, entropy comes from inferences and abduction rather than the new.
If an artist just thinks of something completely new not inspired by others it only appeals to a small minority. It is a minor portion of human creativity, and avoided by perhaps most humans.
It was very unhappy about doing so; I had to stress that it was just for research in the prompt to avoid it flat out refusing to write self-modifying code. But it worked!
"AI," i.e. a glorified chat bot can simulate an FTP server. Something with a well known perhaps almost 60 year old specification.
I am sorry but so what?
This is stupid, expected, and boring.
We can do text, images, audio. What you want, smell?
I would LOVE to read the gpt version of Disney plots.