🤼 Personas

[hejny]

Personas are an abstraction over many of the technical things you need to think about when working with large language models.

• Choosing the right model (according to capabilities, (human) languages, privacy,...)
• Set temperature and top_t
• Writing a system message
• When to continue with which chat
• Adversary checking between personas 👮 Agent adversary expectations #39
• When person does not know, he/she can escalate to another (similar concept as Apple Intelligence)
• Persona can be user itself (this can mitigate PROMPT DIALOGS or/and some async queue to manual process)
• Persona can be one GPT assistant
• Anomaly detection for each persona 🔴 Anomaly detection #40

Synatax

- PERSON Joe, 33 years old, skilled Typescript programmer who uses technically advanced terms. 

Create a new one:

- PERSON Jane, experienced copywriter who writes perfect copy in French and keeps private information in secret

And then just refers to the same person and continues the chat with "him":

- PERSON Joe

But allow you to override any technical settings:

- PERSON Jane
- MODEL NAME `gpt-4o`

[hejny]

🔎 Existing solutions

(Do some research / discussion)

• Has anyone else had the summilar thing?
• Has a project come up with the solution
• Is there a research paper about it?
• Is there an article, video, podcast about it?
• Is this even a good idea?

[hejny]

Has anyone else had the summilar thing?

Yes, DSPy framework