is it? i would write a sentence like this on my own. i guess informed by the thousands (or hundreds, at least) of PMs I have read or contributed to. the same material LLMs would have trained on.
How do people even write anything anymore? I would be so afraid of writing something that someone would suggest I wrote with AI. Reason being, I guess, is that I'm just so surrounded with it and US corporate speak that I rarely notice it anymore.
Agreed, this is some proper nice tinkering writeup that we get far too rarely now.
Lovely project! I'm a software guy who in recent years does lots of CAD for hobby projects (mainly robotics) and orders custom machined parts (lots of sheet metal construction, occasionally milled parts) along with 3D printing.
I find parametric modelling very zen. Stacking operations is very Lego-like, like stringing up pure functions. Plus I can listen to podcasts while I model, but not while I write code - it engages the brain differently.
Now that LLMs are sapping some of the joy out of programming (I use the tools, they're productive, achieving goals and delivering user value is still satisfying, etc. - but the act of writing code is just more enjoyable than prompting, so it's a tad dispiriting that it's getting harder to jusitify) I also find that I get a lot of satisfaction from doing something with my hands. In some ways it's a safer space for technical creativity.
Paragraphica is a lensless, sensorless camera that, when you press the shutter, compiles a bunch of data with GPS, location, time of day etc, and feeds it to an image generator to create the image.
As large language models (LLMs) transition from research prototypes to production systems, practitioners often need reliable methods to verify that model outputs satisfy required constraints. While sampling-based estimates provide an intuition of model behavior, they offer no sound guarantees. We present BEAVER, the first practical framework for computing deterministic, sound probability bounds on LLM constraint satisfaction. Given any prefix-closed semantic constraint, BEAVER systematically explores the generation space using novel token trie and frontier data structures, maintaining provably sound bounds at every iteration. We formalize the verification problem, prove soundness of our approach, and evaluate BEAVER on correctness verification, privacy verification and secure code generation tasks across multiple state of the art LLMs. BEAVER achieves 6 to 8 times tighter probability bounds and identifies 3 to 4 times more high risk instances compared to baseline methods under identical computational budgets, enabling precise characterization and risk assessment that loose bounds or empirical evaluation cannot provide.
