On the whole, these exercises have made me a lot less concerned about losing my ability to make a living. I continue to have lots of other concerns about how this type of tech will influence our societies but I'm trying to take this one existential crisis at a time.

The biggest limiting factor right now seems to be rate limiting. These models are going to have to become significantly more efficient before we can reasonably rely on them to this degree.

So a lot of these steps aren't yielding very screenshot-worthy results as they're establishing systems like item creation and AI. I'm making a decent amount of progress but getting rate-limited like crazy. I also haven't even bothered running tests for the past few steps I'm more interested in […]

Phase 2.4: Field of View and Visibility

(A few phases later since they were mostly prepping for this stage and weren't much to look at). We've got FOV!

Phase 1.5: Water and River Generation
Supposedly we have river generation now... I'm not convinced.

Phase 1.4: Basic Perlin Noise Integration

Alright, this is starting to get pretty wild. It implemented multiple procgen algorithms and added debug buttons so each of them can be tested. Neato!

Phase 1.3: Improved Map Generation with Rooms

Phase 1.2: Basic Random Map Generation
Still plenty of failing tests but the game itself _does_ run and seems to behave as expected.

Phase 1.1: Grid System Implementation. We've got walls and some test obstacles. I actually don't even know what that yellow triangle is...

After asking Aider to use the `Graphics` class to draw directly to the canvas instead of using `Text` objects, performance is _way_ better. It looks a little different but I'm not mad about it:

It looks like this might have something to do with what Phaser is doing under the hood when you make updates to text objects. (Our renderer implementation is using lots of `Text` objects to draw ASCII characters). It _works_ but it's super slow, even on a 10x10 map. Let's see if Sonnet can help […]

Looks like something about Sonnet's implementation is extremely slow at drawing our map (even though it's only 40 x 25). Before I continue, I think I should probably figure that out. Digging...

Also... what?

At this phase, tests are beginning to fail and Sonnet needs to be prompted quite a few times to get them fixed.

Another issue that keeps cropping up is that, as rate limiting occurs, Aider will just die early after trying the API a few times. It becomes necessary to ask the model whether it needs to do anything else to wrap up the current phase of development. This appears to allow it to finish up but […]

Phase 0.4: Game State Management. There's some cool stuff going on here but something you notice right away is that the scene takes ~3 seconds to boot and when it does, the lag between keyboard input and updates to the screen is _extremely_ long. We're […]

[Original post on social.atommac.com]