Nice post, I'll try a few of those in my own file. From my side, one thing in the troubleshooting section that I think is missing is telling the agent that it should collect some proof of what it thinks is wrong before trying to proceed with a fix. I have burnt through a large number of tokens in the past in situations where Claude took a look at the dodgy code (that it had written) and went 'aha! I know what the problem is here' before proceeding to make things worse. Telling Claude to add in debug print statements can be remarkably effective but I'm sure it can also come up with other approaches.
Nothing quite like "I see what the problem is", and then seeing Claude start reading a bunch of files and strategizing the re-architecture of a feature just resolve its own 3-line blunder.
If you happen catch it and you're quick to "esc" and just tell it to find a simpler solution, it's surprisingly great at reconsidering, resolving the issue simply, and picking up where it left off before the blunder.
yes, that is why even in full automated mode, it is better to pay attention.
Sometimes it even tries to git checkout the file when it is stuck with some i indentation issues.
I just never run it in full auto, I look at its proposed changes and more often than not ask it to do it another way. Sometimes I'm just so disappointed that I just go code it up myself.
Congratulations for completing and shipping your game! It looks good. Also an interesting idea to provide the source as DLC. I hope that the sales go well and now best of luck for your next game. :)
My own photogrammetry pipeline because I'm fascinated by the tech (automatically create 3d models from photos). There are also a huge number of commercial applications but I haven't addressed one well enough yet.
So far I've built a first pass of the pipeline using C++/CUDA and used it to power a SaaS and desktop photogrammetry app (free for personal non-commercial use). Got some useful feedback from the initial release of the desktop app back in January and I'm hoping to spend some time iterating to improve further later in the year (currently contracting to generate some funds).
It's possible that some deep learning generative AI network will take over all 3d model generation from photos tasks in the future but I'm hoping/betting that a) classical techniques will give higher resolution, more accurate results for a while yet and b) even if deep learning matches in accuracy and resolution it will always be possible to get better efficiency for big chunks of the pipeline using classical techniques.
I would love if someone would make a photometry app that ingests video (such as iPhone video of my house or drone footage of my property) and outputs a 3D gaussian splatting model.
I want it just for fun, but I’m sure it to real estate agents!
Well, you can just extract the frames from the video and save them as PNGs. I am not sure though if iPhone videos contain the metadata necessary for photogrammetry model generation (camera angle, etc.).
Congrats on releasing this. Having played through and reached the target word, my suggestion would be to try to find a way to lower the difficulty for new players. With Wordle as a new player you can just type in words make some progress and then work out more advanced strategies later on. With Spindle for me it felt frustratingly hard and not much fun at times I'm afraid.
After checking back here a few times I eventually got that I could only spin valid words but they were hard to find so I ended up relying heavily on 2 letter words to reach my final score of 35. Assuming that you support valid Scrabble 2 letter words https://scrabblewordfinder.org/two-letter-scrabble-words you end up spinning words that sound like gibberish most of the time so it seems a bit arbitrary when you can't spin titles like Mr or Ms. :P. Perhaps allow a few free spins to help people unblock the board? Or allow people to skip past problem letters every now and then so that they can spin longer words?
Ah Sopwith! I've got such good memories of playing this in the early 90s. Just 16 CGA colors but the gameplay is so good! :D
10 years later, in the early 2000s I made a fan sequel of Sopwith called 'Camel' https://sopwithcamel.sourceforge.net/ that helped me to get a job at EA. Long time ago now but I've just checked it and surprisingly it still works. Hats off to Windows backwards compatibility there. Anyway, please give Sopwith a go if you've missed it over the last ~40 years it's such a great game. As mentioned in another comment, the authentic experience is to play it on a 386 and then have to hit the turbo button to slow it down... :)
