Posted by Alexander Power
Received: 2025-04-22 18:09:16
Channel: LLM - Large Language Model discussion
To write a few paragraphs on a topic, there is (roughly) a four-step process:
- Choose a topic
- Create an argument
- Choose the tone and style
- Write the words
The machine is, in many situations, better than I am at the task of writing the words. 💡 ( it still struggles with a few tones) It is always about 20 times faster than me.
However, it struggles with the first three tasks.
- The "agent" framework isn't focused on self-willed agents. This is (probably) a good thing. But "fiduciary" agents aren't happening either yet. These solve some of the problem related to motivation.
- The machine prefers to create a good-sounding argument to a logically-sound one. 🔥 ( of course, many humans do the same thing) .
- The machine defaults to a tone that has been over-used to the point of annoyance. It is obsequious and aggressively cheerful. In the context of "a chat-bot for people unfamiliar with the technology", this is a defensible choice. But, most of the time, you need your own tone. 💡 ( It has been long enough that the training data allows "write a tone prompt for this query" kind-of works.)