So I'm trying to create an ai assistant a la Jarvis. Prolly too ambitious but I think it'll be fun. I would like to know is there a way to create an ai that holds continuity (such as reminders, alarms, certain important dates) , and is there a way for an ai to provide real time information such as news or weather? The kind of end goal is to throw it onto a raspberry pi and have it working similarly to anazon Alexa but with more personality. Also, is there a way for the ai to remember preferences?
#LLM continuity or changing information?
woven mesa
silent path
If you can teach a small language model to use tools, you can probably make it do these sorts of things. For instance, an alarm is just calling a tool that sets up a reminder. For preferences, I imagine it would be a bit more complicated but it depends on how you imagine it to work. If you explicitly tell the AI to remember specific preference values, it could also work using tool use.
woven mesa
If you can teach a small language model to use tools, you can probably make it d...
do you know how I could go about teaching it to use these tools?
silent path
do you know how I could go about teaching it to use these tools?
You could try creating a dataset by using a model such as GPT-4o through the API
woven mesa
My goal would be to host locally onto a raspbery pi, would i be able to train a model on my personal PC (quite beefy) end then run it through a raspberry?
silent path
My goal would be to host locally onto a raspbery pi, would i be able to train a ...
I don't think a raspberry pi would be enough for this, but I do not know much about them so maybe I am wrong. I imagine you would have to run the model on a beefy computer, and then setup an API for the RPi to use.
If you can fine tune a sufficiently small model and quantize it, you might be able to run it on a RPi but I am not sure if it would be sufficiently fast for your use.
void swift
Hi before thinking to solve this problem using AI, did you given thought of simple API calls like
- news
- weather
these we can get by scheduled jobs using simple API call
- reminder
- alarm
- important dates
for these you can create simple CRUD app and can write scripts to remind you if that date is today
Also if you want to structure the response coming from API into natural LLM response I would recommend to use Groq API for the same cuz they have enough daily rate limit for personal use.
woven mesa
so you think I should use an API instead of a personal computer trained for it?