Comment by dmonitor
2 days ago
The thing about tools is that they need to be predictable. I can't remember the source, but it's a concept I read that really stuck with me. A predictable tool can be used skillfully and accurately because the user can anticipate how it works and deploy it effectively. It will always be aligned with the user intent because the user decides how and when it is used.
A tool that constantly adapts to how it is used will frequently be misaligned with user intent. Language models constantly change their own behavior based on the specific phrasing you gave it, the context you deployed it in, and the inherent randomness in token generation. Its capacity to be used as a tool will be inherently limited by this unpredictability.
Sounds to me like you're not describing its limited usefulness in general, you're describing its limited usefulness when as considered a tool.
That might seem like a distinction without a point but people are not tools and people are useful. If someone argues people are tools then they don't fit your definition any more than an LLM does.
LLMs are as useful as people I would say, and as unpredictable. I run a small consulting company and we develop custom software for businesses, government, whomever. I often have to interface with other teams and other people. To get the most out of working with people you have to know how to talk with them.
Also, I noticed you didn't say anything negative about LLMs :) You just compared it w/ tooling. So I'm curious, do you think outside of the unsuitability of consideration as a tool, do you find LLMs useful?
It's difficult for me to judge their usefulness as a tool. I've used them in lieu of a google search and gotten good information, but i've also got bad information from them. The lack of reliability means I can't just ask an LLM and keep rolling, but the high probability that it will save me some time scouring the internet means there's some value there. I'm not sure I can provide a definitive judgement on them.