He originally released v1 of the game a few months ago and got great feedback from this community. A big thank you for the feedback, it led to some great conversations and provided a ton of motivation for him to keep moving forward.
v2 was built using HTML, JS, CSS. All written by hand in VSCode. No frameworks, no build steps. He made all the graphics himself and also recorded all the audio.
Note that both parents are programmers so he has lots of hints and guidance along the way. He also leverages Google Gemini to answer coding questions (syntax, how to do something, etc), but the LLM isn’t coding for him and it isn’t available to him directly in his editor.
For the blog post, we talked about the target audience and came up with an outline. He then dictated his blog post directly into Google Docs. Finally, we went through a few rounds of feedback/edits (for more clarity, more words, etc).
I love the new features! I also appreciate the details of what software you used to create the game. My kids have been building stuff in Scratch and I want to have something available for when they’re ready for the next step, so it’s nice to have all these ideas for what they can use.
I'm curious what are the key differences between "a reasoning model" and good old CoT prompting. Is there any reason to believe that the fundamental limitations of prompting don't apply to "reasoning models"? (hallucinations, plainly wrong output, bias towards to training data mean etc.)
The level of sophistication for CoT model varies. "good old CoT prompting" is you hoping the model generates some reasoning tokens prior to the final answer. When it did, the answers tended to be better for certain class of problems. But you had no control over what type of reasoning tokes it was generating. There were hypothesis that just having a <pause> tokens in between generated better answers as it allowed n+1 steps to generate an answer over n. I would consider Meta's "continuous chain of thought" to be on the other end of "good old CoT prompting" where they are passing back the next tokens from the latent space back to the model getting a "BHF" like effect. Who knows what's happening with O3 and Anthropics O3 like models..
The problems you mentioned is very broad and not limited to prompting. Reasoning models tend to outperform older models on math problems. So I'd assume it does reduce hallucination on certain class of problems.
Kids dad here. There's no doubt he had a ton of guidance - both parents are experienced programmers. Many things needed explaining to a 9yo, the same things that would have needed explaining to any new developer.
We went through MANY iterations (test-play/code-review + feedback + dev) before it was released to the public, which meant there was a lot of discussions and lots of opportunities for him to correct many small issues.
Some thoughts:
* DOCTYPE & meta utf-8 - he learned from Khan academy
* meta viewport - I showed him how to test for mobile and pushed to make mobile a priority
* const/let/backticks - he uses prettier in VSCode, which does this automatically
* code modularization - as a result of discussions around maintainability
Improving clarity around charging is on his plan. Its a big issue with the current version.
I'll show him your feedback about the end-game. He was just excited to use document.write()'s side effect of blanking the page if it was the first use of the function. Partly he just wanted to do it because I told him we don't use that function anymore.
