Show HN: I built a toy music controller for my 5yo with a coding agent
github.comThe HN community may find the context of the prompts, organized by each turn in each session, the most useful. See the website/docs/prompts.md and session-X.md files. I also started exploring some workflows for the LLM to execute, organized in the website/docs/tasks/ folder. I found it pretty handy to have the LLM document our work as we went and simply embedded the static site into the executable, along with all the music and logic.
The whole project took me about a day for the backend. The C++ controller itself took only a few turns.
I enjoyed focusing on my son's experience and letting the agent handle the C++, Javascript, and Go code.
I'm still getting started with coding agents, so please do share any tips or tricks to help me with similar projects. I'm most interested in how to work effectively with the agent, like what you see in dev-loop.sh
For those that love the idea of this kind of child-friendly media consumption but maybe don’t have the time, consider Yoto. You can make your own cards that can contain one track or playlists of mp3s that you drag and drop onto a web interface. The yoto then downloads and stores those files, playing them whenever that card is inserted. You can also use the ipad app to browse and play the same content.
Yeah Yoto is our family’s iPad alternative to avoid exposing the kids to too much screen time. On car trips or when they’re just being wild we break them out and the kids love trading story cards and then zoning out and listening. Highly recommended.
>The HN community may find the context of the prompts, organized by each turn in each session, the most useful.
So cool! I'll review fully later but was curious as something caught my attention, do you need to say please and thanks in your prompts for better outputs or is this just anthropomorphism taking over?
I do the same! To me it's a semi sarcastic hedge against the robots uprising and Roko's basilisk.
But more seriously I remember reading somewhere the LLM produce better output with question starting with please, supplementary politeness was not improving results. Probably because the training corpus include many samples where politeness in request produce better response. That apply to the original GPT-3.5, how it applies to newer models your guess is as good or better then mine...
The thanks are unnecessary but I guess they are useful to reinforce his son politeness habits.
It’s anthropomorphism taking over. I pretend I’m talking with a colleague when I do this sort of thing.
The C++ code is bad. Do you have any experience with that language? Why did you decide to use it as is?
I don’t. It works, reliably enough that my son hasn’t had any issues using it for a week. What’s bad about it?
if it works, it works
I know. But it's not a good advertisement for all those AI agents.
Seems like a great advertisement if OP got something working with relatively little time and effort. What's specifically bad about the code?
Great way to allow your kid to play the Frozen song on repeat every hour of the day :D
That was the first album on it for sure and yes he repeats it constantly. Luckily the speaker is in his room.
He’s got a yoto too but won’t wear the headphones, so this has been a nice compromise.
In my experience there is no way to avoid this short of eliminating every trace of copper and silicon from your home.
To be fair, if they manage to play the Frozen song on a classic Game boy color, I'd let them.
I love the user story descriptions.