is there anyway I can try this for myself? that is take the instructions and get the in quotes compiler to output the game
I'd imagine so, OpenAI has a playground and API on their site.
Building games through natural language using OpenAI’s code-DaVinci model
11–20 of 33 posts
Re: Building games through natural language using OpenAI’s code-DaVinci model
#12Read the first page. Seem interesting but not interest in shooting game.
Given that game of life is … can it generate that abd sone if the patterns.
Also can it play go, chess, bridge … etc.
If not is it inherent or just this model.
Not a game developer and hence just question.
Re: Building games through natural language using OpenAI’s code-DaVinci model
#13This - along with GPT - are great ways to create originality detectors, something desperately needed.
The generators get all the attention, but we should be finding ways to use these as discriminators, so that we can find innovative and original projects.
I would love to get a list of Github repos or Steam games ranked on originality/chronologically. Things that are innovative within their own time. There are people making fascinating things, but it takes days, weeks, months to comb through the wreckage to find them.
I have no faith that these models will ever write Slaves to Armok 1 or Finnegans Wake or Dead Stars or original works in their own time - but I think detecting them might be within reach, which is far more useful currently (or at least within my lifespan).
I also think that human programming languages look cool for a demo - but ultimately, there should be programming languages that neatly interface with NNs or whatever - rather than pure text manipulation. I'm sure a lot of resources get sucked up into that alone, modeling syntax, etc. There needs to be a programming language that AI would use, probably directly manipulating an AST of sorts (unless I misunderstood this model, and it's already doing that).
Re: Building games through natural language using OpenAI’s code-DaVinci model
#14Interesting how you still need to have some intuitive sense of what's going on under the hood. (You can't say "make Zelda," you have to ask for an array of symbols and manipulate them.) In that sense it feels like this is still programming, but at a higher level of abstraction with a weird fuzzy compiler. Now we can go from natural language -> JavaScript -> assembly etc. rather than just the last two. Mediocre progra…
Yes.
Could AI generated code mean the death of coding?
I'm wondering if, ultimately, you can get rid of the language as a part that you think about at all. Why not allow the AI to create a language that best suits it? Perhaps this would be hard to read for a human, but who cares? In fact, this would be a good thing for whoever owns the AI.
The issue then will become - as you say - to represent the problem well at a higher level of abstraction. Representation of the problem and knowing what a 'right' answer should be.
Re: Building games through natural language using OpenAI’s code-DaVinci model
#15I'd be curious to see what the upper limit of this is. Could it for example, be trained to optimize video games? I think of the magic fast inverse square root optimization in Quake that dramatically reduced the cost of calculating angles.[1] I bet there's all sorts of non-intuitive optimizations one could do in modern video games that are otherwise too tedious for most programmers to perform. [1] https://en.wikipedia…
> Could it for example, be trained to optimize video games?
In a sense, it already is; nVidia's DLSS [0] and AMD's FidelityFX are AI-driven technologies that allow games to be rendered at a faster, lower resolution, then using AI / ML technology to upscale it to HD or 4K resolutions without upscaling artifacts; the technology fills in the blanks based on a lower resolution frame. Apparently applying the AI upscaling is faster than rendering at full resolution.
[0] https://www.nvidia.com/nl-nl/geforce/technologies/dlss/
[1] https://www.amd.com/en/technologies/fidelityfx-super-resolut...
Re: Building games through natural language using OpenAI’s code-DaVinci model
#16This is one of those things that's so incredible and mind-blowing I really want to share it with friends or family, but WHY it is so impressive is locked behind a high enough sophistication that it would mostly be lost on them. Having written a script a decade ago about a future in which software issues would be solved not by debugging or programming, but by finding the right way to communicate concepts to AIs, it's…
I am still greatly disappointed by the insistence on end-to-end black-box models.
In the end, as impressive as these results are, they are fundamentally trending in the wrong direction. All the benefits and certainty (e.g. security, correctness, and reproducibility) provided by theorem provers and model-driven systems are thrown out the window in favour of fast but potentially wrong or insecure results.
The worst part of this development is the psychological aspect - humans have a tendency to rely on machine generated results and view those as superior. The disconnect between working code and correct or secure code respectively is going to widen using this approach.
A glaring example is found in the blog post: the image manipulation example (7.) contains an error that the author failed to even recognise or mention. Instead of turning the uploaded image into a mosaic as intended, the generated code simply creates a fixed-size black-and-white checker-board pattern. This is clearly neither a mosaic nor image manipulation.
It is a very impressive tech demo, but generating actual software that can be trusted and rigorously checked against requirements will end up using a formal description (i.e. programming language, theorems, or modelling akin to UML) anyway.
Re: Building games through natural language using OpenAI’s code-DaVinci model
#17Interesting how you still need to have some intuitive sense of what's going on under the hood. (You can't say "make Zelda," you have to ask for an array of symbols and manipulate them.) In that sense it feels like this is still programming, but at a higher level of abstraction with a weird fuzzy compiler. Now we can go from natural language -> JavaScript -> assembly etc. rather than just the last two. Mediocre progra…
Yes. Could AI generated code mean the death of coding? I'm wondering if, ultimately, you can get rid of the language as a part that you think about at all. Why not allow the AI to create a language that best suits it? Perhaps this would be hard to read for a human, but who cares? In fact, this would be a good thing for whoever owns the AI. The issue then will become - as you say - to represent the problem well at a h…
> Could AI generated code mean the death of coding?
A rose by any other name...
"Coding" is the formalisation of ideas, algorithms, requirements, and constrains. This task is and will continue to be challenging. Whether you use "prompt engineering" or a formalised language doesn't matter all that much.
I'd be more interested to see whether such black-box model can solve programming problems like returning the best (by some criteria) N items from a read-only medium using a limited amount of resources (e.g. x amount of RAM and t milliseconds).
Given the immense amount of training data it's hard to distinguish a clever search method with some mixing and matching (i.e. copy-paste-programming) from general problem solving abilities.
Re: Building games through natural language using OpenAI’s code-DaVinci model
#18is there anyway I can try this for myself? that is take the instructions and get the in quotes compiler to output the game
It seems to be generated using this:
https://beta.openai.com/codex-javascript-sandbox
Re: Building games through natural language using OpenAI’s code-DaVinci model
#19This is one of those things that's so incredible and mind-blowing I really want to share it with friends or family, but WHY it is so impressive is locked behind a high enough sophistication that it would mostly be lost on them. Having written a script a decade ago about a future in which software issues would be solved not by debugging or programming, but by finding the right way to communicate concepts to AIs, it's…
I am still greatly disappointed by the insistence on end-to-end black-box models. In the end, as impressive as these results are, they are fundamentally trending in the wrong direction. All the benefits and certainty (e.g. security, correctness, and reproducibility) provided by theorem provers and model-driven systems are thrown out the window in favour of fast but potentially wrong or insecure results. The worst par…
>in favor of fast but potentially wrong or insecure results.
Formal methods don't eliminate wrong or insecure results. Formal methods tell you that a program matches the specification when certain conditions are true eg. no bit flips, the computer does not crash, the kernel doesn't kill your process, that allocations succeed, that your program can make progress (the kernel can decide to never schedule your program), etc. You can have bugs in writing the specification where the specification does not match your intention. Even your intention of a system may have vulnerabilities in it. If your specification can't generate code, your code might not match the specification.
Using formal methods slows you down compared to things like testing which can get us most of the way there in less time. Systems can be designed to be robust such that if a machine fails the system keeps on running. If a end-to-end black-box model can get you most of the way there with a sufficiently low number of bugs it may be worth it to use. Time is a limited resource and being 100% correct is not necessarily better than being 99% correct and having extra features.
>The disconnect between working code and correct or secure code respectively is going to widen using this approach.
Not really. People are not going to just start ignoring bugs when they run into them because the software they are using happened to be machine generated.
>Instead of turning the uploaded image into a mosaic as intended, the generated code simply creates a fixed-size black-and-white checker-board pattern
It worked fine for an image I just tried. Just like the prompt it "convert[ed] the image to a 32x32 mosaic." There was no checkerboard, but it may be worth noting that it converted transparent pixels to black.
Re: Building games through natural language using OpenAI’s code-DaVinci model
#20Interesting how you still need to have some intuitive sense of what's going on under the hood. (You can't say "make Zelda," you have to ask for an array of symbols and manipulate them.) In that sense it feels like this is still programming, but at a higher level of abstraction with a weird fuzzy compiler. Now we can go from natural language -> JavaScript -> assembly etc. rather than just the last two. Mediocre progra…
> Mediocre programmers use APIs, while good programmers know what's behind the curtain and can debug them. I suspect this will stay the same, no matter how many layers of abstraction we add. The skill of both such categories (API developers and developers who use API's) is defined by the ability to know the _least_ amount of complexity needed for a given set of requirements. You may be appealing to some "deeper" sens…
> Mediocre programmers use APIs, while good programmers know what's behind the curtain and can debug them.
Personally I'd like to not know what's behind the curtain until I have to. Which category does that put me in?