Post #4507765

another vibecoding experiment: mqtt-senses I'll be doing a write-up about the experience soon. This is part of my project to explain how AI engineering works and what it takes to get quality code from slop. The final goal is to build an LLM server with a full voice and reasoning stack to explain how it all works and, finally, to measure how a gamer-tier GPU stacks up to cloud-scale AI for realistic applications. My goal isn't to normalize the use of AI, but to demonstrate what is realistically possible using these tools. Not only that, but to show that there's a difference between asking a tool to do a job for you and using a tool to complete a task. Obviously, delegating your thinking to AI is harmful, but that doesn't mean that AI cannot be used productively. Sure, maybe AI accelerates laziness, but it can also accelerate other things. Obviously, we will also be discussing the implications of AI under Capitalism. a couple of fun gems from this project: I found a slightly more complete port of hardwood for chicken 5. I'm using hardwood for all my concurrency needs. Though it should be noted that deepseek v4 flash is pretty bad at writing well-formed concurrent code even within the erlang-style actor model. I'm using a handful of eggs that I've been playing with as part of 2682. Namely clojurian and callable-data-structures. I think more work is needed to make data structures that are as ergonomic as clojure using scheme primitives, but its been interesting to see all the permutations that just a couple of small macros can tidy up.