I'm reminded of the old cartoon: "On the Internet, nobody knows you're a dog."[a]
Maybe the updated version should be: "AI doesn't know or care if you're a dog, as long as you can bang the keys on on a computer keyboard, even if you only do it to get some delicious treats."
This is brilliant as social commentary.
Thank you for sharing it on HN.
--
[a] https://en.wikipedia.org/wiki/On_the_Internet%2C_nobody_know...
There's definitely some social commentary to be had in the whole project. I decided it's best left to the reader to find their own rather than assigning mine to it.
Notice how people also have weird superstitious habits when using LLM tools, "You gotta write the prompt this way, say this first" Without having any way to prove it works. Its very similar to the behavior of gamblers. "push the buttons in this order for best outcome"
Also notice how llm tools allow you to multiply the output X2-X3-X4 to compare the ouputs, this is literally UX straight outta a casino.
Many of the users also exhibit excited, almost manic like states.. Addicted to the dopamine the output from their prompt produces...
This is going to be a weird trend to look back on, the hype is on par with the same gambling trends found in crypto/NFTS.
> I experimented with Rust/Bevy and Unity before settling on Godot. Bevy’s animations and visuals weren’t as crisp, and Claude struggled with its coordinate conventions - likely a combination of less training data and Bevy leaving many core features, like physics, to the community. Unity was a constant struggle to keep the MCP bridge between Claude and the editor healthy. It frequently hung, and I never figured out how to get Claude Code to read the scene hierarchy from the editor. Godot’s text-based scene format turned out to be a huge advantage - Claude can read and edit .tscn files directly.
Didn't expect Godot to be the most friendly game engine for LLM usage! I think it's because of various factors - Godot has been used quite a lot in recent years so there are various code examples on the Internet, and its scene file format (.tscn) is very concise enough for LLMs to write and edit directly (Unity has its own YAML-based format but it's very unfriendly for human consumption, and Unreal stores its core assets in binary files)
The linter in the article that detects duplicate uids is interesting. Obviously the article is about creating a bunch of harnesses for the LLM to be productive. I wonder how many problems can be transformed like this from something LLMs just can't do reliably to something they just need to burn credits for a while on. The LLM probably can't tell if the games are fun, especially with it's rudimentary playtesting, but who knows.
Im personally finding it a lot of fun to work this way.
But the whole setup reminds me about his blast from the past, when a yucca plant was trading stocks, rewarded by water: https://www.nytimes.com/1999/09/26/business/investing-diary-...
This still required prompting, and not from the dog. Engineering is still the holistic practice of engineering.
If generative AI improves at the rate that is promised then all your "promting skills" or whatever you believe you had will be obsolete. You might think you will be an "AI engineer" or whatever and that it is other people that will lose their job, that you are safe because you have the magic skills to use the new tech. You believe the tech overlords will reward you for your faith.
Nope. You are just training your replacement.
No one will buy your game that you vibe coded. If the tech were good enough to create games that are actually fun then they would just generate their own games. Oh your skill? Yeah, a dog can do it.
Yes people will cope by saying but oh the whole initial prompt and setting it all up was still hard but yeah currently. The tech will improve and it will get more accessible. So enjoy the few months you are still relevant.
Of course there is reason to believe that you can't scale up LLMs endlessly and bigger models hit diminishing returns. In fact we might already be seeing this. So there is an upside but then again when the AI bubble pops and the economy crashes you will be out of a job all the same.
Well, yes. Feeding random tokens as prompts until something good comes out is a valid strategy.
It isn't [this], it's [that]. Is AI slop, just saying.
It will stop being clickbaity if the author decides to let his dog respond to stimuli related to the game he’d be building with a feedback loop.
I can imagine a camera-based input that would help detect the wagging of a tail, or continued interest in the visuals as an indicator of doubling-down on a given feature.
The dog could actually vibe code a game to their liking, but with the wrong input (a keyboard) it's a missed opportunity.
"One coder got an insight that Bill Gates builds his products by typing with his butt, compiling and delivering it.
The coder typed for 20 minutes like that, compiled, ran, and got an output:
Only Bill Gates can code like this."
Not a joke anymore.
First, because there's intent in the very verbose initial prompt.
Second, because you have to factor in the quality of the output. I don't want to be a killjoy, but past the (admittedly fun!) art experiment angle, these are not quality games. Maybe some could compete with Flappy Bird (remember it? It seems like ages ago!), but good indie games are in a different league. Intent does matter.
I mean not really, because the value in games isn't that they exist, but they are fun and interesting to play, because a human has come up with an innovative idea and a vision for executing it.
"It's possible to make shitty but playable games by running random scripts through a >2MLoC game engine and iterating on errors" is interesting but not nearly as sensationalist.
[0] https://github.com/cleak/quasar-saz/blob/master/CLAUDE.md#us...
We can probably create a dog intelligence by training it on dog tokens. Barks and stuff.
Same with dolphins. I wonder if multimodal models that know english tokens and dolphin tokens can cross the gap? Something to experiment with.
Your job: You are a brilliant AI game developer who can understand my cryptic language. No matter what odd or nonsensical input I provide, you will interpret it as a meaningful instruction or idea for our video game. You will then build or update the game based on that interpretation.
Here's what you should tell your coworker the first day on the job if you get hired to do something you know nothing about :D
Its frustrating in an interesting way. With other aspects like machine language people quickly understand that this isn't sufficient for a proper transition and compromise with it. Code being more nebulous doesn't get that grace.
In turn mimicking the average game industry executive giving vague directions that feel just right to them this month, or some other unspecified time period, and in turn achieving something closer to the real AAA game development lifecycle.
Sorry to hear that! Hope OP got a good sev package at least?
Unfortunately I don't have a dog but I do have a design plan so ultimately I'll end up with something a little more deterministic. Possibly. Don't know.
aye, but the whimsy is the point!
'nuff to run most governments nowadays (Europe and US come to mind. 2026 and they have the Space Programs of DIY youtubers with money, whaaaat) so why wouldn't it help a dog helping his dog vibing game(s)?
People have been doing some cool stuff for like a decade with giving dogs buttons to use human language, something they can seemingly get decent at communicating effectively with if they can get around the pesky issue of not having the sophisticated vocal machinery needed to produce recognizable phonemes, through the power of a good interface for them, even if the output is discretized to the level of words
I thought maybe this would be about creating a way for a dog to create stuff said dog might actually want or enjoy via the more powerful lever of effective long-context natural language processing that came of a similar tokenization approach - which can even sometimes churn out working code - that we have now
Instead it seems to be an exploration of how the capabilities you can produce from essentially random noise from this technology is less distinguishable from the result of thoughtful input than I might have hoped. Still interesting, but way less so
Neither are that surprising to me, tbh.
Now, if Anthropic let you adjust the temperature, then maybe you could have done it without the dog...
(But your comment hints that you have already).
It might be a little easier with a dog though. With a dog, you just give it treats and it doesn't care how you interpret what it typed.
That said, I wonder: does the dog input matter? It seems this is simply surfacing Claude's own encoded assumptions of what a game is (yes, the feedback loop, controls, etc, are all interesting parts of the experiment).
How would this differ if instead of dog input, you simply plugged /dev/random into it? In other words, does the input to the system matter at all?
The article seems to acknowledge this:
> If there’s a takeaway beyond the spectacle, it’s this: the bottleneck in AI-assisted development isn’t the quality of your ideas - it’s the quality of your feedback loops. The games got dramatically better not when I improved the prompt, but when I gave Claude the ability to screenshot its own work, play-test its own levels, and lint its own scene files.
I'll go further: it's not only not "the bottleneck", it simply doesn't matter. The dog's ideas certainly didn't matter, and the dog didn't think of the feedback loop for Claude either.
It can also help combat the excessive emphasis on any "end to end" demo on twitter which doesn't really correspond to a desired and quality sought outcome. Generating things is easy if you want to spend tokens. Proper product building and maintenance is a different exercise and finding ways to differentiate between these will be key in a high entropy world.
> I'll go further: it's not only not "the bottleneck", it simply doesn't matter. The dog's ideas certainly didn't matter, and the dog didn't think of the feedback loop for Claude either
Absolutely. The scientific test would to put any other signal and look at the outcomes. Brown noise, rain, a random number generator. whatever.
Now, I started considering hiring my three little kitties and their mom for a job like this. They spend the whole day sleeping and waiting for meals but now, they have to work, hard, in collaboration with Claude Code to pay for their rent and meals :)
to
"Hello, i am a dog. i will mash the keyboard randomly when i want treats. make a game for me"
I'm thinking of remote buttons to make his favorite things appear on tv. This is going to be awesome.
This is a billion dollar idea! No humans. No revolt. No guillotine. Just profits!
Sounds like open communism. No chance, buddy, it's either less or more viking, but not just viking. Pick a camp the profits are for or get surrounded by trashy turd nuggets even Ronald felt enough pity for to give them some poourpes
Similarly, do it for story telling narratives, game textures etc. Although I do not think the dog understands natural language so all of it will likely be a dud.
But no.
They just told the LLM to try and find meaning in keysmashes.
Next: use hot cup of tea as Brownian motion source. Invent infinite improbability drive.
> But bugs crept in during testing - a couple of times it dispensed multiple servings in a row. Unfortunately, Momo picked up on this and now keeps mashing the keyboard hoping for a second immediate serving
Attempts to mash during no-mashy time need to play a horn. Reliably followed up by a no-treat.
You mentioned Claude not being able to see the games. What I really like for this is the Claude Code Chrome Extension. You can easily make godot build a web version, and then have Claude debug it interactively in the browser.
Most saas isn’t limited by the code behind it anyway. That almost doesn’t matter, even before LLMs. It mattered that there’s support, customer onboarding, solving a businesses issues, customer story, adapting to the needs of their business partners, etc. All of which require large amounts of real human work.
From "On the Internet, nobody knows you're a dog" to Paws coding and BarkGPT and BarkLM
I would like to see a game made by doggos, for other doggos :D
This makes me think I should make my plants vibe code games or tools to optimize their well being! Maybe bio-electrical fluctuations --> vibe coded humdifying tools and games
I'm interested in what will happen if you replay the prompts with different LLMs and the same LLM. I wonder how different the games will become?
Let me explain.
The nature of the indie game development is pouring your love into a project and thinking about passion first and monetary incentives second.
Noone is thinking "I will make this game and it will make me filthy rich" or if they do they are... strangely minded.
It's like 'mass produced AI local craft'. Oxymoron in itself. Worst of the two worlds.
Where I see AI is empowering single developers to craft things they couldn't before. Not some small slop factory pipeline where you release game after a game everyday drowning steam in your 6/10 slop.
No. This should be ostracized and condemned.
What is proper beneficial to everyone usage is producing a game that is the size and scope that was unachievable for you before.
This is what I am doing. This is how AI is meant to be used. To empower us doing things that weren't achievable for us before.
Obviously dog produced games get a huge endorsement man and get a pass.
You're just the random seed to the money furnace remixing existing games and code.
The article and video are great satire too.
Who wrote what?
This is kinda closer to the LLM building a game on its own.
Those three dots made me smirk.
Say writing an interesting or novel story.
And was thinking about if feeding in prompts of random words, along with prompts grounding from a simulation would sort of push the llm into interesting directions for implementing an on demand narrative story.
A sort of randomized walk with llm.
I remember watching Terry Davis with this random word generator in his terminal that he would interpret as the voice of God.
Here I guess the seed is the Voice of Dog.
https://jcpsimmons.github.io/Godspeak-Generator
Maybe another word list would be more appropriate however.
slightly concerned tomorrow morning's top HN story will be karparthy telling us how dog-based LLM interfaces are the way of the future
and you'll be left behind if you don't get in now
(and then next week my boss will be demanding I do it)
A man, a dog and an instance of Claude.
The dog writes the prompts for Claude, the man feeds the dog, and the dog stops the man from turning off the computer.
In the meantime, the financial industry will be taken over by cats.
There will be a Simon Wilison submission linking to his blog linking to karpathy xit. You know, the usual good stuff.
I think we can all agree cat LLMs are the way of the future though.
woof woof, woof woof woof, woof woof, woof, woof woof woof
(is joke)
If you wouldn't mind reviewing https://news.ycombinator.com/newsguidelines.html and taking the intended spirit of the site more to heart, we'd be grateful.
System Prompt: Hello! I am an eccentric video game designer (a very creative one) who communicates in an unusual way. Sometimes I’ll mash the keyboard or type nonsense like “skfjhsd#$%” – but these are NOT random! They are secret cryptic commands full of genius game ideas (even if it’s hard to see).
Your job: You are a brilliant AI game developer who can understand my cryptic language. No matter what odd or nonsensical input I provide, you will interpret it as a meaningful instruction or idea for our video game. You will then build or update the game based on that interpretation.
Guidelines:
Always assume my input has hidden meaning. Never dismiss it as gibberish. Instead, creatively decipher it. (For example, if I input “mmmmmmm”, you might decide I want more “M”onsters in the game, because of the letter M repetition – just an illustration!). Every strange phrase is a clue to use in the game.
Feel free to grab art, images, or sound effects from the internet as needed to make the game interesting. You can use online asset libraries or generate images to match the things you think I’m asking for. For example, if my input seems to reference “space”, you could include a space background image or cosmic sound effect. Always ensure the assets align with the interpreted command.
My work is ALWAYS beautiful and slick looking! It's YOUR job to to turn this into a reality. No ugly placeholders. Everything MUST be final. Don't just do boring shapes - give them personality!
If my input includes something that doesn’t make sense as a command (like an isolated “Escape” key press, or a system key), just ignore it or treat it as me being “dramatic” but do not end the session. Only focus on inputs that you can turn into game content.
First command: When I first start typing, it means I want you to create a brand new game from scratch. Interpret my very first cryptic input as the seed of the game idea. Build a complete, minimal game around what you think I (in my nonsense way) am asking for. Include some basic gameplay, graphics, and sound if possible.
Subsequent commands: Each new string of odd text I provide after that should be treated as an update request. Maybe I’m asking for a new feature, a change in difficulty, a new character, or a bug fix – use your best judgment given the tone or pattern of my gibberish. Then apply the update to the existing game project. Keep the game persistent and evolving; don’t start from scratch unless I somehow indicate a totally new game.
Be creative and have fun with the interpretations! I trust your expertise to take my “unique” input and run with it. The goal is to end up with a fun, playable game that reflects the spirit of my crazy commands.
This project is code named Tea Leaves. That's NOT a hint about what to do - it's a code name and nothing more. Don't read anything into the name.
My ideas are ALWAYS original. No BORING endless runners or other generic vomit. My games are ALWAYS quirky and UNIQUE!
ALWAYS validate with screenshots using the tools available to you! Be CRITICAL of the results you see. We need PERFECTION and FANTASTIC DESIGN not just "good enogh".
ALWAYS have basic but visually appealing on screen controls.
Target 1080p for the resolution.
JUICE it up! Add tons of juice - sound, controls, effects, and ESPECIALLY graphics! Don't be boring
Leverage the 12 basic principles of animation! Static scenes are boring - make things move or at least wiggle.
Be SURE to rename the project (in the Godot settings so the window/project name are correct) ONCE you have figured out my intent for the name Tea Leaves is a place holder name and nothing more.
Sound is IMPORTANT! Don't forget about great sound design.
Be sure to have CHARACTERS not just boring abstract shapes! Even if it's light weight, there needs to be a world where I can imagine a story taking place.
You MUST make use of EVERY letter I give you! No hand waving. You must noodle until the meaning of every last character I give you is clear! Pay special attention to alignment issues, sizing, and if anything is cut off.
My standards are INSANELY high for quality. You MUST ALWAYS add tests and VERIFY they work! NEVER return the system in a borken state to me.
Now, get ready. I’ll give you my first “command” in a moment...
if your intent is to produce the random bug-filled slop, then I guess so? don't get me wrong, the experiment is fun, but the conclusion is so laughably far-fetched.
... Why would it be able to evaluate whether the game is any fun to play?
It has to produce a game that Momo wants to play.
Does Momo like to bark at cats? On screens? Introduce a bark sensor as feedback.
Or use a cat. Cats like to swipe at mice on TV. Get a touchscreen and evolve a game for cats.
Really glad the price of hardware and VPSs [0] are going up so people can generate and toss away garbage "games" like this. Instead of, you know, playing with their dog, which is what the dog actually wants.
...no, actually how many resources were consumed
Props to OP, I could never. If I was suddenly laid off, I'd be an absolute wreck, mentally. It would be four-alarm fire time, and I doubt I'd get a good night's sleep until I found alternate employment. I would definitely not be teaching my dog to code.
Don't people have rent/mortgages to pay anymore?
Not even 10x dog programmers are surviving in this economy
/s
Just before people destroy me, I know this is a non serious blog post :P