I’ve felt a growing concern that the more we rely on LLM’s to do even problems that are far more efficiently done with lesser technologies, the more energy-wasteful we will be. A very basic machine-learning algorithm can bound image objects with boxes, for example, and has been able to for years.
Maybe LLMs will be the new scripting language. You technically can do everything in a scripting language, you just don’t want to, especially when performance is a consideration. Most people prototype things in scripting languages and write a real implementation in a language with less overhead and better verification tooling. We might end up doing the same with LLMs.
with the difference that, while “traditional” programming is incremental, when you want to eg change LLM or implement its suface behaviour in code you’re none the wiser.
I’ve felt a growing concern that the more we rely on LLM’s to do even problems that are far more efficiently done with lesser technologies, the more energy-wasteful we will be. A very basic machine-learning algorithm can bound image objects with boxes, for example, and has been able to for years.
This is a neat demo though!
Maybe LLMs will be the new scripting language. You technically can do everything in a scripting language, you just don’t want to, especially when performance is a consideration. Most people prototype things in scripting languages and write a real implementation in a language with less overhead and better verification tooling. We might end up doing the same with LLMs.
Exactly my thoughts. It’s another tool to help making MVPs faster.
with the difference that, while “traditional” programming is incremental, when you want to eg change LLM or implement its suface behaviour in code you’re none the wiser.