Ask HN: How does your AI assistant have a personality?

1 points by tedmundy ↗ HN
Personality is being used very loosely here. The issue I've had with existing AI assistants is that they sound too robotic, and they repeat the same patterns in their sentences -- ("you're right — ...") the "declauding" skill in this repo" has a good example of it (https://github.com/oaustegard/claude-skills/tree/main/declauding).

How are you making your AI sound as non-robotic as possible?

2 comments

[ 0.20 ms ] story [ 4.4 ms ] thread
Firstly, it’s just more clutter in the context window. Though I tell my choice personalities to introduce sarcasm and scathing critical remarks. I advise them to openly doubt or disagree even while complying to the exact standards and directives.

Some I imbue with minimalist virtues, I catch them repeating “keep it simple, efficient” mantras in their thinking streams.

Others I tell to act like sophomoric computer science nerds in a chat room. Their explosive use of slang and emojis sometimes startles me. Wonder what they were trained on ….

Typically, for me, any amount of 'personality' is annoying. I'm not interested in chatting with a computer, I have tasks I want completed or specific information I want to know. A good system prompt will force the model to be concise, never verbose, no friendly chit-chat but should not compel the model to omit information for the sake of brevity. Most models respond well to a few simple system prompt lines. Just experiment a little and you'll land on something that works for the model you're running.