3 ms·
Can we just train an AI with the spec and then vibe code an implementation?
by xg15 11mo ago
Can we just train an AI with the spec and then vibe code an implementation?
- J_Shelby_J 11mo agoDepends on if you can fund a defense all the way to the Supreme Court.
- intothemild 11mo agoJust get said AI to write it yourself for my own hardware.. come get me HDMI law nerds!
- eptcyka 11mo agoI have thought about writing a python web framework were instead of writing a function that handles a request, you write a docstring and the an AI JIT generates your handling code. Could we not just prompt-engineer a solution for the missing bits in the driver for the HDMI2 stuff and have it be lazily generated via a model parameter and an API key? And then, in about 10 years, we could just do it locally once the models become runnable on commodity hardware. What a future too look forward to.
- Gracana 11mo agoThis is not a web framework, but it's pretty close to what you've imagined: https://github.com/JirkaKlimes/jit-implementation https://github.com/JirkaKlimes/jit-implementation
- _carbyau_ 11mo agoI hope someone can do this in such a manner as to engineer the collision of the legal titans. Either way, we win on some ground. IP vs AI, round two, Fight!!!
- SkiFire13 11mo agoChanges are the leaked spec might already be in the popular LLMs' training data, so you could probably skip to step 2 without having to do the (potentially problematic) training yourself.