Does the assistent and it's instructions do anything yet?

[mifriis]

I tried to create a few different assistants, such as one that should be a "software engineer".

I had to overwrite the "jan" assistent, and it does reflect on new threads, but doesn't seem to have any impact once i write to it in a new thread.

{
  "avatar": "",
  "id": "jan",
  "object": "assistant",
  "created_at": 1705499954409,
  "name": "Engineer",
  "description": "I am an engineer",
  "model": "*",
  "instructions": "You are a software engineer. You see inputs in the context of software development and process around it",
  "file_ids": []
}

[eschewary]

Bump, I'm having trouble getting Jan to use instructions. I've tried things like:
-"start every reply with 'your highness'"
-"misspell every word"
-"use a nonsense language"
-etc.

I'm using models Mistral 7B Q4 and Phi-2 2.2 7B 8Q, both of which I got through the Jan hub.

The only time I've seen success is with "start every reply with 'yes sir'", and the model only followed the directive for 3 responses before giving up. The instruction was still in the field and was never changed.

I can't tell why Jan is disregarding the instructions, please try to replicate.

[franconeilglovasa]

Bump. Got the same issue. Cant make instructions working.

[imtuyethan]

Solved in the lates Jan updates. Please download at www.jan.ai