“Build and ship a computer game” is on many a programmer’s bucket list, mine included. However, I am quite cognizant of the reality that most people who claim “if only you could make game X with Y” don’t know what they’re talking about, because even if they are a programmer, they haven’t worked in the game industry or even in consumer software. Yours truly is guilty on all counts.
While I am far from the first to be pushing the frontier on using LLMs to
attempt to generate games, it does seem that 5.6 Sol at xhigh effort is at a
threshold where it can work on ambitious feature sets within a single prompt
without collapsing under the weight of its own slop, if you cast the right
spells steer the model through one small chunk at a time and pay down tech
debt as you go with regular codebase audits.
Games seem particularly interesting because quite frankly, instead of the typical priorities you’d need in B2B-SaaS land, “game feel” is king and whatever your code looks like is secondary. I’d assume intimate knowledge of a codebase usually goes hand-in-hand with honing game feel, but what if you can just run enough experiments with a cheap model once you’re mostly feature-complete? Of course, you need to have the wisdom to know that the change you’re tweaking isn’t gonna be a major refactor…
Perhaps this is just a bout of psychosis, but I have been really, really impressed by 5.6 Sol. I had been collecting notes on Inverse Kinematics and combined them with notes on other locomotion papers like this one and it produced an IK system that plays quite well with Box3D.
I think I want to FAFO. But don’t get it twisted, the point here isn’t to say “I’m making a game engine”, but it is to say “if video games are all about feel, then a vibed engine should be perfectly suitable for experiments in game feel, no?”. I’m not gonna run towards anything too groundbreaking, but over the course of the past few days, I have been very, very impressed with how much I could get done with my own engine cobbled together from open-source C++ libraries.