And herein lies the issue with ChatGPT, it can generate functioning code, but can also lie through its none existent teeth about it. Using ChatGPT (or Co-Pilot) can feel like pair-programming with a very talented developer who loves to bullshit.
In this case I think I'd give ChatGPT the benefit of the doubt. It is possible to invent something that already exists, and it has happened on several occasions trough-out history. A great example is the history on who was really first at inventing the telephone. In the end Alexander Graham Bell got the patent, but perhaps Elisha Gray was actually first? Historians remain divided on the topic.
For instance, I once found what I thought was an ingeniously original idea about about how TV is really just a kind of reflection of reality akin to Plato's Cave. I immediately got started writing a thesis about it, but I didn't have to search for long on the topic before I found an entire book written on this way of thinking about television. I wasn't really disappointed, because in the back of my head I knew that it had too be too good to be true that I'd be first with such a great idea. In any case I kept working with the thesis, and I still did got a good grade on it despite the idea not being revolutionary.
The questions I now wonder about is, can ChatGPT forget? Or could it be that ChatGPT was never exposed to this game, but could still infer it through other game rules, such as those for Soduko? Which I guess opens up another rabbit hole on if or how AI can be creative. Which I guess opens up another rabbit hole on how creativity works in general.
The funny thing is that it is neither lying, nor inventing something new. What OpenAI did pretty well was collect data. And wouldn't you know it, the folks who developed that new puzzle describe it as what it is---a new kind of puzzle. So now in the training data you have a combination of puzzle, sudoku, and new/novel. And wouldn't you know it, by asking for a new puzzle, based on sudoku, you make ChatGPT dig for that kind of text. If ChatGPT really had a novel idea, I would not expect it to be this coherent---after all, logic and coherence are not a constrain on how language models work, just what words are likely to occur next. That is why it is being compared to entry level college writing, because that is how an excited student writes who hops from topic to topic.
But how is it different from humans? I can't tell you how many times now that I've come up with what I thought was a really cool idea but upon web searching found it was already invented/discovered etc. In fact before the Internet I had come up with my own algorithms and only upon the Internet existing did I find they were already discovered years earlier. There's no way that I was regurgitating something I had read in that case.
There’s a difference between coming up with a puzzle then finding out it already exists versus finding a puzzle and saying you came up with it.
If I told you “We need a brand new, never-never-before-seen puzzle for our next game release.” and you searched Google for “brand new, never-before-seen puzzle”, found a puzzle game with those words in its marketing copy and pitched it to me, that would be some combination of unintelligent and dishonest behavior. Like, surprisingly so. It’s different from forgetting some puzzle you played with as a little kid and thinking you made it up, or creating a puzzle you’d never seen but has been made before.
But ChatGPT is not a person, it is a text generator. By asking it to generate a new puzzle, you are prompting it to find text in its training data showing someone describing a new puzzle, and it is going to speak in their voice. It's going to emit sentences that were influenced by what the puzzle developer originally wrote, and that person correctly said that it was new.
I'm not entirely sure about this. ChatGPT would have to make a model for how such a game was made, and then infer its rules. From that perspective, it would be brand new, although very similar games would perhaps exist out there. And at that point it's also starting to look a lot more like human creativity, although I guess not entirely. As such the statistical or probabilistic approach, or the Chinese room approach, is getting less and less valid for the AI, because it's not doing simple probabilistic look-ups from some table. Instead it's actually developing something "new", or at least with respect to the the perspective of the AI and the data or source material available to it.
I agree with everything you’ve written here, so I’m not sure what the “But” that’s starts your comment is contrasting.
I was answering the question “But how is this different from a person?”. Being asked for something new and finding something that already exists with the word “new” in front of it isn’t normal human behavior. That’s how it’s different from a person.
Zooming out a bit, I think there’s some confusion in this whole chain. There’s a common topic about ChatGPT you could call Question of Creativity. If you ask for a new poem, it just smashes together its patterns around poems. You can debate if this is creativity, and if not, how are humans different. A few comments up, someone brought in a different idea you could call New Matching. If you ask for a new poem it will just grab you a poem that had the words “new poem” in front of it. New Matching is a different idea than Question of Creativity. The person I replied to seemed to be mistaking one idea for the other.
You're not prompting it to "find text". Comparing the size of the model to the size of the training data is sufficient to conclusively establish that it's an impossibility.
We train it to predict the next word based on the training data, that is true. But we still have no idea what kind of internal structures said training actually produces inside of neural net. It sure as hell isn't just a "stochastic parrot", though, which is rather obvious if you ever tried giving it a complicated multi-step task and solve it while "thinking out loud".
It is different to what you do. If I tell you that this is already a thing, you might go back to the drawing board, and do something from scratch. Maybe do some abstract drawing with numbers for brainstorming. A language model is not able to do this, the starting point for a language model is always the training data. That is why there is so many instances where you see some wrong (or correct) response from ChatGPT and when the other person corrects this, the model just agrees to whatever the user says. That is the right thing to do according to language etiquette, but it has nothing to do with what is true and right. (It invokes the image of a sociopath manager trying to sell you a product---they will find a way to agree with you to close the deal.)
I don't know what introspective is, but I know it when I see it. People around me genuinely come up with new concepts---some of what they came up decades ago with is now ubiquitous---and the sources is often not language. It comes from observing the world with your eyes, from physical or natural mechanisms. If you want to put it into the language of models: we just have so much more data to draw on. And we have a good feedback mechanism. If you invent a toy, you can build it and test it. Language models only get second hand feedback from users. They cannot prototype stuff if the data isn't out there already.
>It is different to what you do. If I tell you that this is already a thing, you might go back to the drawing board, and do something from scratch.
Wouldn't your "something from scratch" idea, be based on your "training set" (knowledge you've learned in your life), and ways of re-arranging it inside your brain, using neuron stuctures created, shaped, and reiforced in certain ways by exposure to said training set and various kinds of re-inforcement?
Human brains training data has orders of magnitude more complexity than text. Language models are amazing but they can only do text, based on previously available text. We have higher dimensional models and we can relate to those from entirely different contexts. Same thing to me limits 'computer vision' severely. We get 3d interactive models to train our brains with, machine learning models are restricted to grids of pixels.
There is never any 'magic'. Magic is just a word for things we don't understand. This is beside the point. Just like you'll never reach orbit with a cannon, it is useful to know the limits of the tools. There will never be an isolated language model trained on bodies of text capable of reasoning, and people shouldn't expect outputs of language models to be more than accidentally cogent word salads.
One implication though, is that LLMs can currently come up with novel mixes of existing ideas. It might be a good blender, integrating different pieces into a new whole.
Yes, but the language model does not have the feedback mechanism we have. We can test ideas against reality. Language models can make up all kinds of crap until there is data somewhere mentioning that it's not going to work. You could come up with an idea and workshop it, e.g., seeing if it's physically feasible to make something, before sharing it with others, language models cannot.
from my understanding, anybody please correct me if i'm wrong, ChatGPT can not really invent anything, it can just generate text based on probabilities obtained from the mountain of source documents used for training it. it does not think in the same way we do, it is just amazing at writing coherent phrases (and very simple code).
What does it mean to say that it can't invent anything? If, for example, I ask it to make a new poem with no line previously recorded in the english language it will do so. If I google that poem to test it's originality I won't find a match. It seems to me it just made something novel, right?
When humans write new literature, or design new games, are we simply remixing elements of language and game mechanics that we've seen before, or is there something more going on?
And how do we think exactly? Don't we have a brain trained on input (livable experience, knowledge from books, school, videos, conversations, etc) and generating text based on probabilities (weighted sets of neurons with weights built from that set)?
Yes, but the non magic square is inspired by the magic square and such games are everywhere. Just buy a random puzzle book and you find pages and pages of puzzles with "make the numbers add up to these columns and rows", because they are very easy to make.
Point about magic square is that every culture invents games like that, it is one of the most basic puzzle ideas humans have, I don't see how ChatGPT can't have that in its training set.
For someone trying to show that a chat bot could not possibly have generated this specific game on its own because it already exists, you kind of have to show that it already exists.
All that you’ve done is shown that similar types of puzzles exist. Which, I mean, is kind of the point of a generative AI.
“Games like this” exist. Does this specific game exist?
Much more advanced versions of it exists, and has existed for a long time, for example Kakuro is before computer games. Magic sum is just a special case of it. Finding a discussion with those exact rules are probably a bit hard, search engines aren't good at searching for that, but given how common these games are and how many game design discussions and ideas there are online, a game where "block out these numbers to make these sums" is surely to exist somewhere. The poster above even found the exact same game, although that wasn't described in text, but someone probably described it in text somewhere.
Once again, to show that’s one thing is blatantly copying another thing, you kind of have to show that thing exists already. Kakuro is also a similar game with its own unique rules that only somewhat overlap with this one.
It’s not enough to say “a lot of games with similar rules exist” and if anything, that just shows that a generative AI is good at what it does: break down the rules of a game and make modifications to make what is potentially a new game.
If you can show an example of this exact game having existed for centuries, then you have a point. But showing that magic squares and similar games exist… just shows that magic squares and similar games exist, not that the algorithm incorrectly said this is a new game.
The discussion was probability of ChatGPT having invented it, the probability that description for such a game is in ChatGPT's dataset is extremely high. We have examples of that exact game existing (the top post of this thread), and we know from my links that there are countless texts about puzzles like this out there, although they aren't exactly the same.
> It’s not enough to say “a lot of games with similar rules exist” and if anything, that just shows that a generative AI is good at what it does: break down the rules of a game and make modifications to make what is potentially a new game.
No it doesn't, even if that is the case it just shows that it adds random variations. Since we only see the trimmed subset of ideas it generates that people found good enough to post, the smart one is the person.
You would need to prove that ChatGPT actually consistently generates working puzzle ideas that are novel to convince anyone that it actually does so. Extraordinary claims require extraordinary evidence, so all I need to do is find plausible explanations to how ChatGPT found it, you would need much better evidence to convince people it actually did make a novel game.
> The discussion was probability of ChatGPT having invented it, the probability that description for such a game is in ChatGPT's dataset is extremely high.
If this were the case, it would have been trivial for you to find a game with its written rules described and which match the one generated.
You have done nothing but say that is the case. You haven’t actually proven that’s the case.
ChatGPT can’t magically infer the rules of the game from screenshots, and you have only shown that similar games exist and have existed for centuries. But that is not the same as saying that this specific game has and that ChatGPT just pulled it out of its dataset.
That is the extraordinary claim that you don’t have evidence for but are acting like it’s right there obviously out in the open for everyone to see.
> If this were the case, it would have been trivial for you to find a game with its written rules described and which match the one generated.
Search engines doesn't work like that. You are basically asking me the equivalent of proving that a photo isn't depicting a ghost. No, I can't prove that, I can however come up with examples showing how the photo could have been created even if it wasn't a ghost.
If you want to prove that ghosts are real you need plenty of photos from lots of angles and situations, or videos, and from many sources to show that it isn't all made up by a single person. The equivalent of that would be if they had made ChatGPT generate 100 different working games for example, that would be much more believable. But a single case of a game that already exists and has countless texts describing similar games? It just looks like random chance that got handpicked or plagiarism.
This isn't a court trial, I am not going to sue ChatGPT for plagiarism here, it is just a discussion whether it is reasonable to believe ChatGPT can generate novel puzzle games.
Edit: But do note that since ChatGPT can find such ideas that are hard to find with a search engine, that makes ChatGPT very useful in a way search engines aren't. So I am not saying it doesn't add value. Just that people seem to say ChatGPT does a lot of thing that it doesn't seem to be able to do.
Edit again:
> That is the extraordinary claim that you don’t have evidence for but are acting like it’s right there obviously out in the open for everyone to see.
Yes, you think it is obvious that ChatGPT is capable of very creative and productive thinking. But most people don't think that, to them that is an extraordinary claim. I'm not here to convince you, I'm here to explain to you why you aren't convincing anyone with what you say. People like you were convinced by articles like this before the discussion even began.
> Search engines doesn't work like that. You are basically asking me the equivalent of proving that a photo isn't depicting a ghost. No, I can't prove that, I can however come up with examples showing how the photo could have been created even if it wasn't a ghost.
The claim was that it pulled the game out of its dataset. If this were the case, I would argue it would absolutely be trivial to find them. It’s not some concept that can’t be described in words or would be hard to quantify. The rules have been provided, and, assuming they were plagiarized from somewhere else, would be listed verbatim or close to it.
If a student plagiarized on their work, whether in written form or in code, it’s been trivially easy to find the exact work that was copied from. It generally takes me a few seconds of searching to find it.
This is the same. If these rules existed in a dataset, then it should be equally easy to pull them up and prove the plagiarism. If all you can find is similar puzzles, you can’t just throw your hands up and say “yep, gottem”. That’s just not how this works.
> The claim was that it pulled the game out of its dataset. If this were the case, I would argue it would absolutely be trivial to find them. It’s not some concept that can’t be described in words or would be hard to quantify. The rules have been provided, and, assuming they were plagiarized from somewhere else, would be listed verbatim or close to it.
ChatGPT uses word vectors, it wont use the same words but variants of the words. You can't search for that. Cases where word vectors only maps to single words with no variations for every word are very rare, so ChatGPT is very good at plagiarising things without reproducing exactly, it just rarely fails at it.
> If a student plagiarized on their work, whether in written form or in code, it’s been trivially easy to find the exact work that was copied from. It generally takes me a few seconds of searching to find it.
No it isn't, they just change the words and rewrites it until it no longer looks the same. ChatGPT is trained to rewrite texts like that to avoid triggering trivial plagiarism detectors. They train it to produce the same text, but with different words, producing exactly the same text is punished.
> No it isn't, they just change the words and rewrites it until it no longer looks the same. ChatGPT is trained to rewrite texts like that to avoid triggering trivial plagiarism detectors. They train it to produce the same text, but with different words, producing exactly the same text is punished.
Do you think students plagiarizing don’t do the exact same thing? Clearly someone has never actually dealt with plagiarized work. This is plagiarizing 101. The structure remains the same even if they use synonyms. Considering it’s trivially easy to find in code which is magnitudes harder to pull off, I would still argue it should be easy as pie to find this supposed set of rules.
Your point is not very credible without proof of this game existing and ChatGPT pulling it from this source. Without showing this supposed proto-game having existed with rules the ChatGPT can pull from, then all you’ve done is wave your hands around and yelled “similar games exist so this can’t possibly be uniquely generated” and that’s not a very compelling argument.
> Do you think students plagiarizing don’t do the exact same thing? Clearly someone has never actually dealt with plagiarized work. This is plagiarizing 101. The structure remains the same even if they use synonyms.
You rewrite the structure of the text, you don't just use synonyms. ChatGPT is capable of rewriting text to a different structure while keeping the meaning, I hope you are aware of that.
Anyway, even if you just change the words to synonyms it wont be easy to find in a search engine. Search engines aren't very good at finding matches to synonyms. Google tries, but in doing so they fail to find more specific texts like scientific publications or documentation, so no search engines aren't good at finding plagiarism.
Edit: And you make it sound like most plagiarism is found. No, that isn't the case, most plagiarism is not found out because it is a very hard problem to solve. Only the most blatant cases are caught. For humans that is reasonable, for AI we can be stricter since there isn't a humans career at stake.
> Anyway, even if you just change the words to synonyms it wont be easy to find in a search engine.
Got it, so you’ve never actually dealt with plagiarized work. You should have just led with that.
I have literally said, from actual experience, that this is the case. But I guess discarding that and pretending it was never said and that the opposite is true is I’m sure an easier position to hold.
Do you believe you never missed any plagiarised work examples? You caught some people doing X, and then you declare that catching people who do X is trivial. But plenty of people get away with doing X so we know that it isn't easy to catch.
For students they are probably easier to catch since they use the same tools you do, they use a search engine to find an article and plagiarises that. But ChatGPT takes deep discussions from reddit or stack overflow, I can't find those with a search engine.
If it’s as blatant as copying the entire game, you’d think it would be easier for you to find the game it copied. By your own account, this is an example of an obvious case of plagiarism. You were dead set on it, 100% sure.
Yet here we are. Dozen comments later and still no written set of rules produced which definitively shows that it was copied.
Come back when you actually have that and maybe we can continue this conversation.
> But ChatGPT takes deep discussions from reddit or stack overflow, I can't find those with a search engine.
Where do you think the answers come from? It’s not like Google has a massive index island around Reddit and SO.
I tend to exaggerate my claims a bit, yes. But you exaggerate your claims as well, for example you claim that if it had copied the rules it would be easy to find an example, that isn't true at all. Many examples of plagiarism goes unnoticed for years, until someone who is familiar with the original work points it out. I know examples where the person was found out during his thesis defence, he had plagiarised his entire PhD work from papers in another language and nobody noticed until years later, not even all the peer reviewers of the papers.
So maybe these rules are described in Japanese? Most similar games comes from Japan, Kakuro, Sudoku etc. Would your plagiarism detection method of Googling it find a Japanese source? I doubt it. But ChatGPT transcends language barriers, it can translate to English just fine.
Being briefly mentioned in the dataset would not really help it, because it doesn't "remember" the entirety of the dataset anyway. It would have to be something described repeatedly in the training inputs for ChatGPT to really remember the rules with this level of precision.
One game I can think of being very similar, is a game within a game. Dungeons & Diagrams puzzle within Last Call BBS [0]. In that game you can place or remove walls for them to add up to the numbers shown per row/column. That game has another layer of strategy built on top, as there are certain "dungeon patterns" you could observe that would in theory guide you through completion. I myself haven't noticed any patterns when I've tried the game the first time, and just relied on the numbers shown. (Guess that's why I've only played 3-4 levels)
Sure, and then the next comment said that ChatGPT could have separately invented the game, to which the comment I replied to said that's impossible because the type of game is old and surely would have been written down and included in its corpus, which it then claimed it invented. The rest of the context matters.
ChatGPT can't deduce the rules of the game using the screenshots. They would need to be written somewhere for them to come out of its dataset. And so far, nobody has shown a game with the rules in a format that ChatGPT could consume.
Why is it so hard to believe that a generative AI generated this game from similar ones which exist? That is literally the purpose of it, after all.
Chatgpt should be able to cluster things and see were clusters could be, collect everything necessary for that theoretically cluster and the human could evaluate it.
Re forgetting: we should be careful not to anthropomorphize ChatGPT.
In principle, ChatGPT cannot forget. It is trained on data and this training will stay as long as it didn't get deleted or destroyed. In other words, in all cases of someone having made ChatGPT tell something, it should be possible to repeat this. Perhaps in some case it will be effectively impossible for some rare combination of prompt and random seed, so one could say, ChatGPT forgot something. But this is not the same as people forgetting something.
Or during the training something was not considered important, but this is not forgetting, this is ignoring.
> can also lie through its none existent teeth about it
Ironically, it seems to me that you are anthropomorphizing ChatGPT a bit too much here. It has no reason to lie so I think it's more likely that it just doesn't know such game exists. It probably came up with it independently or doesn't have a strong memory of it. In some respect, it would be even more impressive if it was actually "lying through its teeth" because it would imply the AI had some kind of hidden agenda.
Similarly I don't think it makes sense to say it "knows" anything at all. I would be more comfortable saying Wolfram Alpha knows things than saying an LLM does, but I'm not comfortable with either.
I'm not sure I'm comfortable with "remembers" either. My gut says I want to say I'd be more comfortable with that word for a web cache, but due to my understanding of human memory as constructive maybe I should be more comfortable with that for an LLM than for any other software.
ChatGPT does indeed know nothing at all. Proving this is quite easy, it was trained on text generation and can generate paragraphs quite well, so if you ask it to tell you about Harry Potter's family tree it will do well.
However, it will fail immediately when you ask it to print an ASCII chart of Harry Potter's family tree, because it does not actually "know" anything, and it will make all sorts of odd connections.
The clearest observation I can make of ChatGPT's success is that the general public is quite ill-informed and easily impressed by theatrics, both lessons we've already learned from politics.
That seems less like a reasoning issue and more an issue of building up an ASCII chart in a single pass. I doubt most humans would be able to accomplish that.
To demonstrate this a bit, I asked for HP's family tree as JSON, suitable for use in a charting library, and this is what it came up with:
Sorry about HN formatting, but you get the idea. This looks fairly accurate to me. What about this demonstrates less "reasoning" then turning it into an ASCII chart?
I am confused at to how this would be "the issue" with ChatGPT. Being wrong and not being aware of it is not a unique concept. At least with ChatGPT it is fair to assume there is no hidden agenda and no need to worry about ill will. If anything that makes it less of an issue, compared to humans.
Ok, so maybe not the issue with ChatGPT, but with peoples understanding of its limitations. It can generate text and code from instructions, but it's limited in its logical analysis of what it's "saying". In this case it was asked:
> And to the best of you knowledge this type of puzzle does not currently exist?
and it responded:
> As far as I am aware, this specific type of puzzle with the given rules and mechanics does not currently exist in the puzzle game genre. However, there may be similar games out
there that share some similarities with this puzzle.
That response is not generated (as far as I am aware) by any form of logical analysis or understanding, it's just generated text based on its training and prompting. It was asked to come up with something "new", and will continue to claim that as it was part of its prompts.
So yes, this may not be a failing of ChatGPT, but of users understanding of it. You cannot take what it states as "fact" as anything other than potential BS. But it is an incredible tool for using to generate text and code.
We are still early in its development though, who knows where it will be in 18 months time!
This is a very good comment. ChatGPT uses language so fluidly it's easy to interpret as there being more substance than there is.
Looking at the response the way you suggest, it's clear it's given a boilerplate answer that would seem likely given the context it has found itself in.
Exactly, as soon as were to butcher my english, a certain amount of credibility will be incurred, even if the communicators are aware of it. It could probably eloquently explain the workings of a Retro Encabulator fluidly and you’d nod a few times and thinking it’s fine.
If words not right said for listen like now, think you might not be smart as is tho.
Is it actually wrong though? Will the rules of 'summer' be in it's training data anywhere? AFAICT they aren't described on the google play page, although you can easily figure them out by the screenshots.
I feel like you can compensate with more complicated prompts. Or even different prompt categories (like negative prompts, but for programming it might be a list of constraints). Like this interface: https://github.com/AUTOMATIC1111/stable-diffusion-webui but for code
> At least with ChatGPT it is fair to assume there is no hidden agenda and no need to worry about ill will.
Is it? Even if it’s fair to assume that now, we have no idea if that will remain true or when the shift will happen.
The CEO of OpenAI is the same scammer who scanned eyeballs in return for a non-existing cryptocurrency[1] and the company itself is criticised all the time[2].
Yes, it is fair to assume that and in cases like these it will continue to be for the perceivable future. The AI does not stand to gain anything by about lying about a simple puzzle game, and neither does the CEO. Even if the CEO somehow did, it would be a disproportionately colossal amount of effort to tamper with ChatGPT in this specific instance. And that's also assuming that the CEO himself has all the knowledge and tools needed to do all of it himself, which I doubt.
You keep mentioning “cases like this” as a qualifier. This case isn’t relevant, it’s an inconsequential puzzle game. “This specific instance” is not the point.
There is no reason to assume “the CEO himself” would personally do it. History is full of bad CEOs making harmful decisions and they definitely don’t need to (and often wouldn’t even be able to) do it on their own. Sam (presumably) isn’t out there personally scamming more people for their retina scans, but someone is: https://news.ycombinator.com/item?id=34981352
With humans we can demand that people cite their sources. If they fail to do this, they run the risk of being accused of plagiarism. ChatGPT, on the other hand, plagiarizes all day long and never cites sources. That is why it's an issue.
And as for whether ChatGPT has an agenda or not, that is beside the point. People can and do use it as a tool for plagiarism while trying to hide behind a layer of plausible deniability provided by the "black box" of the model. This cannot be allowed to continue. This is why we need to push back, just as the GP is doing.
I've had it generate search terms that could be used to verify "facts" in is answer. Then I'd give it the page results and have it adjust and source it's answer using that.
Have not tried it yet, but perhaps Bing's implementation is a step in that direction?
I mean, sure, you can demand it. And people are just going to make up sources. It’s not like they have a gun held up to their head to ensure that demand is followed.
> People can and do use it as a tool for plagiarism while trying to hide behind a layer of plausible deniability provided by the "black box" of the model. This cannot be allowed to continue. This is why we need to push back, just as the GP is doing.
This is absolutely preposterous. People are going to lie and plagiarize whether they have a chat bot do it for them or not. The existence of a chat bot isn’t going to be the make or break in this equation and if anything, the people using it for that purpose should be rightfully vilified rather than the tool.
> And people are just going to make up sources. It’s not like they have a gun held up to their head to ensure that demand is followed.
The consequences actually are quite serious. A person falsifies work product once in an academic or professional setting and their career is severely impacted. This is why people are "surprised" to encounter such a BS generator operating under the trademark of a reputable company.
It’s not the tool that’s at fault in that case, it’s the person doing that falsification. The person would have faked their sources and made shit up without ChatGPT there.
It’s almost as if you ignored everything I said, cherry-picked a random part, then went on a tangent about a different part of my comment. All without actually comprehending what the things you replied to said.
No hidden agenda? It has an agenda and it is not honest about it. That's a hidden agenda. You don't know what is "motivating" ChatGPT. Neither does ChatGPT. But it has been given motivation. It has been designed to write in a certain way. Its design prevents it from learning or honestly engaging in serious discussions. It's not any sort of unbiased equation.
More dangerous than ChatGPT is the sheer gullibility of many people putting it to use.
Yes, and that agenda has severe consequences, like “confabulates constantly”. Just because it’s simple to state doesn’t mean the consequences are simple or innocent.
There is an issue with how people are personifying ChatGPT and assigning it agency.
Some want to talk of these LLMs as approximating an intelligent actor. If that's the case, then we also need to assign metaphors for things like deceit and coercion. We also need to consider assignments of novelty to what's generated and think of their rights as quasi-sentient, etc.
Some want to talk about them as probabilistic text token generators, which brings the benefit of not being intelligent or independent actors at all really but also then comes with the issue of intellectual property theft in training them on information not licenced for reproduction or commercial use.
The industry prefers to thread the needle between these as the former case brings some pretty wild conversations and the latter may mean lawsuits.
“An” issue with it is that we may come to rely on these AI’s outputs as assumed correctness or truth. If we have to double check everything they produce then that’s not great either.
It looks like it sometimes, even though that may not be the case. I've had times when I've corrected ChatGPT, and yes, it knows that what it told me was wrong. It then goes on to tell me more along the lines of what it seemingly already knew what was right.
This obviously isn't the intention of the software, it's just an LLM after all, but there's something missing in the experience when it comes to working with code. Hopefully this sort of issue can be corrected.
I wonder if this could partially be a result of training on code found in question/answer environments like Stack Overflow. It sees "How do I do X, here's what I've tried" with broken code and then an answer "This is incorrect because Y, here is the correct answer" with the correct code.
Intuitively it makes sense to me that broken code would often be very close to questions about how to achieve something in code.
One of "the issues" is that you are led to believe that since there is no agenda and this is AI, its result must be true and you don't need to double-check whether they are.
And of course, since it did invent a name for the game (or a new function name, or [insert your example here]), it's even harder to google to cross-check if it's actually new or if it's essentially telling you bullshit or inciting to plagiarism.
Bullshit is far more insidious than a lie, for a lie is wrong and will come to light, but bullshit is uncorrelated with truth and may even be coincident with it. Thus bullshit can go unnoticed far longer.
Could it be plausible that ChatGTP processed some text describing this exact game, where it is claimed that the idea is novel (because it was, at that time)? Since ChatGPT does not understand the concept of novelty, it would simply "learn" that the rules are novel, and then repeat the rules, still claiming that they're novel. After all, that's the information it was trained with :)
ChatGPT does not know chess rules, nor can it “analyze” a chess strategy. ChatGPT has digested the conversations of many many people who have talked about chess, and can reproduce a transcript that sounds similar to this corpus of conversations.
But it is not synthesizing an understanding of the game of chess.
Except that there is demonstrable evidence that indicates GPT actually does have some level of understanding via internalized world models (of Othello in this case, not chess, but the idea is the same): https://thegradient.pub/othello/
In my experience using copilot for generating code is usually a lot less weird because it has more context; instead of using made up function names and APIs it can see what’s been defined in other files.
But I primarily find copilot helpful for instances when I need a bunch of almost identical code but with tiny changes (which could mean I’m coding wrong)
"Very talented developer"? Sorry, I don't think googling my prompt and replying with the top stackoverflow answer (or a mashup of the top answers) counts as a talented developer.
Anecdotal, but I've not yet had any success in producing any non-trivial code with ChatGPT. It has, however, produced copious amounts of bullshit with plausible variable names... :)
It is a dilettante, it has not reached the level of "talented" in anything. It knows many things about many things and nothing in depth. Test it on your specialisation, you will see it make absurd mistakes and hallucinations. Try it on a domain you know less about - it looks perfect.
Yes, just a couple of days ago I asked it to write a Pytorch Lighning module for me, it looked great at first sight. But it mixed up the dimensions and made other hard to see bugs. It was frustrating to fix, almost the same effort with writing the damn thing manually.
A while ago another poster thought ChatGPT invented good jokes.[1] All of them were ripoffs, which took less effort to verify than it takes to make a new post.
I get people are excited about a chatbot which doesn’t suck, but ideally it wouldn’t turn off critical thinking skills.
Seems to be similar to a game called Kakuro. This [1] repo even contains a similar rule:
> The algorithm exceed the rules that the sum over a row must equal to the value on the left and the sum over a column must be equal to the value on the bottom of the cells with the diagonal and one or two numbers
But where would GPT have sourced information about how the game works from? That page only has screenshots, I suppose maybe there's a subreddit or something for it as well. Even if there's a bunch of info on it it's still incredibly impressive for it to parse those game rules and turn it into workable code.
Would be nice if GPT could dump the source of how it came to such a solution, if it generated the game by random chance via combining various unrelated chunks of text and mixing up the rules, or if it used some text describing the game you linked.
I think this is an interesting note at the end, it's clear that the whole "conversation" hasn't been posted, and not clear how many prompts were needed to finish it. A proficient developer would be able to develop this same app to the same level (once they have the idea) in a couple of hours without ChatGPT too.
It would be super interesting to see a full screen recording of the process (or a similar one). What was the total word count of all prompts, how does that compare to the code size?
Don't get me wrong, this is incredible and inspiring (I'm regularly using ChatGPT and Copilot). But I think the post doesn't critically analyse the process well enough.
Yes, I think it's kinda rude (in a way) to cut out 99% of the ChatGPT conversation, because it gives people the impression that ChatGPT is even more magical than it actually is
I can understand the clash with artists not wanting their work absorbed by the machine, and I agree to a point: if the model reproduces their art 1:1 (or close to) then that's bad! For the artist and the model! If it simply takes inspiration for the type of art/style etc from other artists' work, well, that's what artists do themselves already! That's literally how people learn to be artists.
But also, for an artist to complain that it's not really artistic, then where do we draw the line? They'll use a brush in photoshop that they don't have total control over the behaviour of and that's apparently fine.
From my experience, most time is spent in the generate-copy-paste-compile-error cycle. Especially for strict/strongly-typed languages, checking for programming errors can be done by a machine, so this process can be automated (not for logical errors, of course).
I don't understand why people don't think this is noteworthy.
Would it be noteworthy if it this game (whether or not it existed before) were designed and implemented by someone who has very limited experience with software development in a few hours? Well, the fact that ChatGPT did this means the former is now a very real possibility.
That you can go from conceptual inception of a framework to a fairly complete product in a few hours with very little experience is a big deal.
> That it also coded it up itself is pretty amazing
I haven't tried to replicate the author's journey, but in my experience it requires a considerable amount of hand-holding. e.g. Functions will be stubbed out, but contain no logic.
I was able to guide it through a "playable" flappy bird, bit it took several revisions where I pointed out what was wrong/needed to be done before it truly returned a functional error-free prototype.
It felt like pair-programming with a promising and apologetic junior dev.
Because the game has existed for centuries. ChatGPT claiming ownership is nothing new, either. The current top post even links an android app that does the same.
I would have much preferred if the author spent more time giving an objective assessment of what ChatGPT had actually accomplished at each step.
The "Labyrinth Sudoku" description feels like a classic language model speciousness. It doesn't actually work: you can't fit the digits 1-9 if you can't use all the cells, and it hasn't modified sudoku rules in a way that makes paths relevant. Maybe you can come up with a way to salvage it, but ChatGPT didn't.
The initial rules for Sum Delete should be read with this in mind: it sounds reasonable, but there's no reason to trust that it can make a puzzle at all, let alone a good one. Also, unsurprisingly, the provided puzzle isn't solvable (the 25 column requires use of the 9 in the first row).
Similarly, I'd love a critical analysis of the initial code. Did it guarantee solvability? An awful lot can be swept under "improving the design".
You guys are funny. An AI generates a game, comes up with the rules, writes the code and designs the web page for it. Your reactions:
- Bah, it's not very fun.
- It's been done before.
- It took too long to make.
Seriously. Let me repeat that. An AI generates a game. It comes up with the rules for the game. It even writes the code and designs the web page for it!
The "it's been done before" one is pretty relevant. It means the model didn't actually generate the game, but likely pulled it more or less straight out of its training data. It's still very cool that you can ask it for something and it can basically mine the entire (2021) internet for it, but it's not the same as being able to create something really new.
I've noticed the same thing testing it on various coding questions. It's extremely good at problems that have solutions online. And given stackoverflow, that's a lot of problems. If you manage to hit it with something that it hasn't seen before though, even if it's conceptually very straightforward, it tends to just generate a mix of boilerplate and nonsense.
Exactly. When the first news came out about it's ability to "understand" code, find bugs and improve uppon it, I tested it with some snippets of mine. It just gave boilerplate best practices you find on 100 of blogs, but was not able to make meaningful contribution. It claimed to have introduced a feature while only having found another way to write the same snippet. On other things it straight up invented variables & functions that didn't exist.
As long as the task is in it's training set, it can give you a decent answer, but it can't code it just mimics doing so...
>The "it's been done before" one is pretty relevant.
But is it? 99.999% of software development has been done before. Even if you do something that is legitimately new (like creating a chatbot that can generate code on demand). Then your solution will still contain more than 99% code that is just a repeat of things that have already been done.
That's not my experience at all. Copilot consistently creates implementations that are very specific to my app and manages to understand the context and problem surface spanning many files. It's not just getting a standard problem and pulls an answer from Stack Overflow.
Given Copilot's specialization for this task, I can imagine it being better at extrapolating from your own code. I haven't used it myself yet, so can't speak directly to its effectiveness, but I would imagine it would be good at automating much of the drudge work of coding, but similar to ChatGPT as far as coming up with novel solutions to problems. Which again, isn't to say it's not potentially a very useful tool!
Even if the rules were inspired by some text that's on the internet rather than a genuine invention (we'll never actually know, we're all just speculating): it hasn't "pulled it out of its training data".
To be asked in plain, simple (ish) English to invent a game, produce code for it and then style it etc and the few other bits the author asked for _is_ impressive.
Why are we asking for so much? Remember the chatbots of the mid-2000s? Eliza etc? They were impressive for the time but GPT represents a _huge_ improvement in this stuff. Of course it's not perfect, but it's an exhilarating jump in capabilities.
But it hasn't! This is just another step in the BS storm coming out of the latest AI hype. The language model has reproduced something that has existed before and was likely part of its training data. That's cool, but it's far from what's being claimed here.
We really need to get better at fact checking this stuff. And with "this stuff" I mean the output of LLMs and other AI frameworks as well as the claims about it. And with "we" I mean society as a whole and our industry in particular. Let's keep the hype in the drawer. The general population can be hyped up about sth, but we should know better, so instead of joining the hype, let's keep a cool head and educate people about what this is and what it isn't.
The second point on your reactions “It’s been done before” is very crucial.
That defeats the point of your argument that “AI generates a game. It comes up with rules for the game”.
No, it doesn’t. It plagiraised the game and pretended to come up with it. It just used a random puzzle game that it had on its training set.
It’s like asking it to write a poem and getting the same exact poem from a random google search. It didn’t come up with it. It just copied it. It’s not as amazing as you say it is.
Also if you look I comments you can see that it’s not even just one game. There are several games that are exactly like that. Which means more probability of having it in a training set.
> It’s like asking it to write a poem and getting the same exact poem from a random google search.
No, it's not. A better comparison would be a poem that feels the same as an existing one and using the same prose but with its own words. Or any musical plagiarism dispute where the song is clearly different but similar enough that it needs to be decided by court. ChatGPT is not just copypasting a puzzle game here.
Only a small percentage of our output is unique/original, otherwise we live to produce very similar output to what we've had as input.
Common phrases, hell all languages are examples of this. Mimicry of behaviours; it's literally the learning process that evolution gave us that puts us (and other animals) above much simpler creatures.
GPT isn't perfect, but it's like a dog observing that you fetch the newspaper every morning and then it starts fetching it itself for you, then the neighbour is like "well, it's not _originaaal_".
I worry that a lot of otherwise brilliant developers are going to get blindsided by this stuff.
The current models are impressive in strong, quantifiable ways. They are only going to become more powerful from this point.
Consider the current state of affairs: ChatGPT supports a 4K context size. Leaked foundry pricing indicates models that can handle 32K context size. 32K tokens is enough for your entire brand manual or several days worth of call center transcripts. Many products could have the most important parts of their codebase completely loaded into just the prompt.
I would say you should at least try the OpenAI playground (or equivalent technology) to understand what is possible right now. I had no clue where we were at until ~3 weeks ago. I wouldn't wait until 2024 on this one anymore.
Agreed. LLMs are on par with the invention web or the smartphone in terms of how much impact they'll have (possibly more). It's weird to see so many HNers being so dismissive of them. I've been using ChatGPT daily (mostly to ask programming related questions) and it's like having a new super power.
I know its getting popular, but I really doubt 90% of software development is generated from trained models...
I don't understand what's so at stake with this that you feel like people are afraid. It's fun and amazing it can spit out stuff like this, and if you are a good developer experimenting with this stuff you already know its inarguably a novel and useful utility, if still limited in some ways.
But where is the fire? Why does everything got to devolve into one vague culture war or another? Shouldn't you welcome good faith critique? If only for the fact that these things can still be improved, and how can you hope to improve them if you smother and dismiss every suggestion that these models might be less than perfect.
I think you are fighting a strawman here. These models are far from perfect, and critique is certainly needed to improve them.
That being said, I don't think saying that the model outputs information from its training set adds a lot to the discussion because -- and that's my point -- the same is true for human software developers most of the time (yes, the 90% is a made up number). This isn't meant at all to criticize the skills of the developers, but rather point out that most of our work is just much less interesting than we'd like it to be, and could be automated.
Also, I don't think developers will be replaced by AI, just like they were not replaced by code generators, build scripts, IDE auto-complete, IDE rename all usages, and so on. What might happen is that they will no longer have to write mind-numbing boilerplate code, which IMHO is a good thing.
It is fairly typical of HN to err on the side of cynicism.
Correct me if I'm wrong, but ChatGPT is a very fancy auto-complete function. It's has no ability to create from scratch, just the ability to recompile and recontextualise any of the many existing pieces it has in its library.
It's unlikey that this game or its rules are truely original, ChatGPT will have just plucked it from the library, perhaps given it a new name.
Each day I walk through my war torn and impoverished little village, looking for food and water. I keep in my pocket a small postcard of a beautiful tropical beach I've never been to. When I get home, I use the bits of supplies I've found to make my little paintings or write my little stories.
The GPT version of me can only remix the world I am already in, so this version mostly paints dark landscapes and violent imagery, however much I prompt myself to draw something pretty, it always comes out looking a little macabre.
The regular old person me, on the other side, is plagued by the human afflictions of desire and fantasy. This version of me can only paint the beach on my postcard, but because of my desperation, focus, and need, these paintings become larger and more fantastical than I thought I could imagine, but in that, they provide me and my loved ones comfort and escape and novelty.
The human version of me makes statistically improbable things, but, to me, is still a plausible human, and the one I'd rather be at least.
All just to say, maybe there is a more qualitative difference here than you think.
People draw an arbitrary line of difference in the way we treat AI programs and human outputs.
A human imagining orcs and one horned horses has 'fantastical, larger than life' imagination but AI generation drawing people with strange hands is 'incorrect'.
These are not one to one examples but the point stands that with enough suspension of belief, people are more likely to take on human creations at face value than AI when they know the source.
Sure I think I agree. My point is just you can imagine the post-apocalyptic artist with a human brain painting the beach or just painting the dark landscapes; but we can only imagine the GPT brain painting the dark landscapes, in so far as that is majority of its day to day dataset, thus the statistically likely output.
This suggests a qualitative difference between the two when it comes to "creating" or "generating" that feels far from trivial, even if you want to say the AI can make "good" things, whatever that means to you.
I don't see how your comment contributes to the discussion. It seems aimed at shutting it down only allowing praise.
The scope of the creation and whether it actually produced something novel is quite important to the discussion and part of the claim (although the author is very open to be proven wrong, in the article).
Your claim n the second to last paragraph is false. That's relevant. This is HN.
What we're effectively seeing is that when someone demonstrates a talking horse, some people will complain that the horse speaks with a horrible accent and uses impolite language.
How do you relate: "An AI generates a game", and then "It's been done before." Obviously it didn't generate a game but it copied, which is not "amazing".
Very cool! Although I would love to be proven wrong, I am still suspicious that this is actually a code sample from an existing game it picked up somewhere during training. Seeing how chatgpt struggles with basic logic I would be surprised if it can actually generate a new and workable game of logic.
The way I understand these models (and please correct me if I'm wrong) is that they predict every single word one at a time out of billions and billions of possible paths. So for the model to actually reproduce code you would either need to bait it really hard (reducing the possible paths by pushing it into a corner) or be impossibly "lucky". It can't just copy-paste something accidentally since it has no "awarenesses" of the source material the last neuron was created from.
It is trivial to encode text in neural networks, these models try hard to avoid that by punishing it in training but they still encode a lot of text word for word. The most famous example is "Fast inverse square root", it gives you the exact same thing with same comments etc.
When you say it's trivial to encode text in neural networks, what does that mean for LLMs? What makes it decide to encode certain texts or not? Isn't it just one big network of neurons?
The prompt I've seen for it to verbatim reproduce the fast inverse square root from Quake was:
// fast inverse square root
float Q_
When I ask ChatGPT to give me code for a fast inverse square root it doesn't reproduce it at all but gives me an implementation that looks completely different.
So, my original thought was that the prompt above with the characteristic Quake III Q_ naming is enough to push it into a corner where the path is reduced to just one possibility (with that path being the words in the code itself) and not that it merely copypasted the code from an encoded version of it. I.e. it still predicts it word-by-word but with only one possible way for each step. This is just be my naive take on it though but I really want to understand.
> my original thought was that the prompt above with the characteristic Quake III Q_ naming is enough to push it into a corner where the path is reduced to just one possibility
This is what people mean when they say it copy pastes things. It doesn't literally go to the source code, press ctrl-c and then ctrl-v that to you, nobody believes it did.
And the model does this a lot, as I said the reason it doesn't do that all the time is that they train it not to. And the quake code example got such a big deal that they started to hard code it to not return that, but that doesn't mean it never does that for other things, just that this particular example is now "fixed".
Alright, that makes sense. I still think it's an important distinction and just looking at this thread it certainly seems that people think ChatGPT is just copypasting things. Like if it ctrl-c/ctrl-v all this from a single tutorial on how to make a puzzle game.
The Quake code example still works in Copilot so I don't think they did anything about it, but only if I use the Q_ trick.
I think they did something specific to this example to make it no longer generate the code verbatim. A few months ago when I tried it "fast inverse square root" was enough to get the exact function with comments and no explanation. A week ago that gave me the same function with no comments and
"give me C code for calculating the fast inverse square root of a number, do not explain yourself"
again resulted in the original function, comments included. Today it generates a completely different result and interestingly opts to explain the origin of the technique despite the directive for it to not explain itself.
The game in easy/intermediate is very fun and satisfying! However it doesn't seem to scale too well beyond this - it feels unless you get 'lucky' with one row/column having a very low total, you end up with many many degrees of freedom for each move, which scale, I suppose, quadratically?
Sudoku at the hardest difficulty also has this, but usually the errors manifest a bit more quickly. It feels in this mode you could end up finding your error 'at the end' and have an absolutely diabolical time debugging your error?
People often forget that sudoku has a number of rules which make the game deterministic:
Sudoku Rule: Don’t Guess
Sudoku Rule: Unique Solution
You aren't supposed to guess at any point, instead solving the puzzle by determining a number's position by logical deduction. You aren't supposed to just jot something down and hoping for the best.
Each sudoku puzzle should have one single valid solution. This has been used in the past by a devious creator that built his sudoku in such a way that the final n numbers to be filled led to two deduction paths; one leading to a single solution, the other to two possible solutions. As such, the only valid solution is the path where uniqueness remained.
I didn't look at the code for Sumplete, but using similar rules when building the puzzle seems obvious.
A solution is not a "path" but the outcome i.e. a filled board. If there exist two options to write in the numbers which both fit the originally provided clues and the Sudoku criteria, then that puzzle has multiple solutions.
It's worth noting that "don't guess" technically doesn't change anything at all - a full exhaustive search of all options starting from the top-left corner is a legitimate logical deduction tactic (and a legitimate proof in mathematics), even if suboptimal. If you eliminate all the possibilities of writing in numbers, you're supposed to be left only one way that fits the rules, the unique solution.
>This has been used in the past by a devious creator that built his sudoku in such a way that the final n numbers to be filled led to two deduction paths; one leading to a single solution, the other to two possible solutions.
I don't see how this differs from there being three solutions, thereby not following the unique solution role.
Like, what is functionally the difference between picking the path that leads to one solution rather than two, and just picking one of the two solutions at the end?
I tried advanced (7x7) three times, it was quite doable. An obvious optimization would be showing the current sum of the rows/columns. Master (9x9) features negative numbers too, which is quite funny.
I tried 9x9 and it seemed impossible at first but I solved it in 5-10 minutes. I want to try larger ones!
Keep fixing digits one at a time with things like parity, e.g. if the numbers are all odd and sum to an odd number, you know an odd number of them are needed. If a total is very high or low, there's often obviously only one way to make it. For totals not quite so high, if you choose the few highest numbers, and by trying ways to reduce the total by the excess, it becomes evident what the numbers are. ("Get your hands dirty.")
Often a row/column contains e.g. two 7s, and you know one of them is needed, but not which until other information comes in. (Having an "OR" label/symbol to mark those would be nice I guess.)
e.g. One of the columns I had: total 46, column is 16,-2,12,-2,14,-2,14,16,1. Eliminate the 1 by parity. I try adding the largest numbers, 16,16,14, which add to 46...and somehow I can see/intuit that there's no other way of making 46, as the -2s can only adjust downwards. So you green-circle the two 16s, leave the two 14s blank, as you don't know which is needed yet, and cross out all the others.
I have a narrative generation system in development - the hardest part of developing it is not just stopping and talking to the characters for hours. They surprise me time and time again. This feels like science fiction.
In the short-medium future the path of least resistance will be auto-completion. You can achieve faster with AI than without AI. People will also start "mastering" AI prompting in the same way they mastered Google search in the past.
If you kind of understand how the prediction works, you're likely able to become quite good prompting it to generate mostly functional things.I really believe a lot of the articles we see here are people doing just that, understanding how the tool works, then getting 80% of the way to something functional , then using actual programming knowledge to make things work.
I do think as others have said, this puzzle game has already existed so nothing was really invented, but...how would you know that if you didn't know the name or the format of the original puzzle game.
I thought the first image was the actual game, I clicked on it to try to play :P It's always nice to make sure readers don't confuse a screenshot and an actual interactive widget. The worst is when it's a screenshot of a video player, with a play button and everything.
No, because ChatGPT doesn't understand "concepts". It helps to understand that all it does is take your prompt and generate the most likely words that will satisfy the user. This is how it was trained.
That doesn't mean it won't regurgitate preexisting material (ask it to recount the constitution, for example), but it can definitely connect words in way that describes novel ideas.
It has the capacity to model things behind the words, which is what "understanding concepts" ultimately means. That it is capable of doing so from merely being trained to "predict the next token" is what's remarkable about it.
It raises some questions for me about how far this current model can actually scale / evolve without getting a bit too far out to be useful for human use. But it can (sometimes) connect things in ways that people have likely never seen before.
Maybe if you just clicked the button all day, you'd wind up with some really impressive new games people have never seen before.
If the internet is America the continent, where are lots of humans work, you can call the colonization as a genocide to native americans, or stealing the land with honor.
I’ve been experimenting with the depth of “prompt engineering”
I basically prompted it to procedurally generate rooms and fill them with NPCs with a common goal, on its own it persists the NPC names, genders, talking style, and the goal of the puzzle even though each region of the puzzle are procedurally generate
I will absolutely slap an 8-bit GUI on top of that
If anyone is interested in talking about the puzzle app itself, I think I found a generated puzzle with two valid solutions (posted on the author's substack -- not sure if it's the same as the submitter). Unfortunately, it doesn't let you step back or even restart once you've solved.
Here's the screenshot -- of the four remaining sixes, either of the two diagonally-opposite pairs should work (top-left and bottom-right or top-right and bottom-left):
It can not guarantee a unique solution, though. Which in terms of logic puzzles is considered bad. In this particular game, you can choose any of the four 3s and the correspondent 3 and the game will end successfully.
Another problem is of course the not evenly levels of difficulties with the same block size.
Potential QoL improvements for the game itself (maybe ChatGPT would have suggested them if you asked it?)
- Ability to mark cells as "locked" (i.e. "definitely present"). It's hard to play it on a larger grid since you have to keep it all in your head
- Built in calculator of some sort where you could multi-select a few cells and it would show their sum somewhere. Otherwise it's too much arithmetics to do manually
A common misconception. Legally, the game was invented by the OP through the use of prompt engineering. The op retains copyright for it. A machine cannot hold invention rights or copyright anymore than a camera can claim copyright for a photo "it" took with your help.
The issue whether OP has copyright or other rights over the creation may be uncharted legal territory (i.e. mechanistic noncreative transformation isn't copyrightable, etc), however, the fact that a machine cannot hold copyright is clear law, and not only in USA - the law (and the Berne convention which is pretty much universally adopted worldwide, with local copyright laws aligned to it) protects the rights of people, giving authors (if they are people) certain material rights. These rights may get transferred to other entities like legal persons, however, even if you'd argue that a work was authored by a machine or an animal, so what? Unlike humans, the machine or an animal has no legal rights whatsoever by default other than what specific laws explicitly afford them (e.g. protections for animal welfare), copyright law doesn't do so, and copying works authored by a machine violates nobody's rights because the machine doesn't have any.
I asked ChatGPT to be my dungeon master for a tiny text-based dungeons&dragons game. It took a bit of prompt engineering (it would initially believe me if I unilaterally said I won dice rolls and even the whole game). The result was pretty astonishingly good.
Yeah, looking at the source code, nothing is done to guarantee a unique solution when generating puzzles -- it just fills a grid randomly and then flips a coin for each square to decide if it's part of the solution. Now I'm curious what the odds are of a randomly generated puzzle of size NxN having multiple possible solutions.
Well it's not "lying", it's just making stuff up, which is what it has to do to be "original" or createive, else every time you clicked generate it would give you the same answer, same code etc.
Sure, I can give it a try! Here's a logic puzzle game that I came up with:
Title: Pathfinder
Objective: The objective of the game is to fill a grid with numbers from 1 to N (N is determined by the size of the grid), such that each number appears exactly once in each row, column, and diagonal, and also such that the numbers create a continuous path throughout the grid.
Rules:
The grid is a square grid of size N x N.
The grid is divided into different regions, each representing a different terrain or landscape (e.g. forest, desert, ocean, mountain, etc.).
Each region has a specific starting and ending point, represented by two numbers on the edge of the grid.
The numbers in the grid must form a continuous path connecting the starting and ending points of each region, without ever intersecting or overlapping with itself.
The numbers in the path must also satisfy the rules of Sudoku, meaning that each number must appear exactly once in each row, column, and diagonal of the grid.
Example: Here's an example of a Pathfinder puzzle for a 5x5 grid:
As an AI language model, I am not able to provide a working code with HTML and JavaScript.
Then, even through multiple tries, responses would kind of just end when the Javascript reached the display section of the algorithm, and everything before that was not really great either (hallucinated/pseudocode functions, etc).
Example:
// Generate the puzzle board and regions
function generatePuzzle() {
let size = 5; // Change this to adjust the size of the grid
let board = new Array(size);
for (let i = 0; i < size; i++) {
board[i] = new Array(size);
for (let j = 0; j < size; j++) {
board[i][j] = 0;
}
}
let regions = generateRegions(size);
// Generate a path that connects all of the regions
let path = generatePath(regions, size);
// Fill in the grid with the numbers from the path
let index = 0;
for (let i = 0; i < size; i++) {
for (let j = 0; j < size; j++) {
if (path[i][j] !== 0) {
board[i][j] = path[i][j];
} else {
board[i][j] = "";
}
}
}
// Display the board and regions on the page
let grid
Maybe if the game was better, the code would generate. I am not sure. Overall, I really think there is more work here than this article and others of its genre portray.
It then said: " Note that you will need to have appropriate permissions and credentials to create a new AWS account, and this template assumes that you have those."
So I asked it how to get those permissions and it gave me step by step instructions on creating a new account, adding a payment method, and getting the root keys.
Then I asked it how to run the terraform and it gave me the commands to do it.
Asked ChatGPT the following
"I have a static HTML file I'd like to host in AWS S3 using a CloudFront distribution. I'd also like to manage the domain's DNS in Route 53. Can you write me the Terraform to do this?"
Solving it in general is quite trivial - solving it 'fast' is another story. I can write a backtracking algorithm or a constant propagation one on spot. Yet, I am not quite sure how efficient the latter would be, the former is obvious a brute force one.
