Recent Posts61
CyberJedi's Ultra Hal Workshop / Re: Smarting up HAL with free LLM model (Ollama) ?
« Last post by knight2000 on September 29, 2025, 01:10:13 pm »@cyberjedi whats UVE?
There are endless possibilities with any of this stuff, just boils down to how creative the developer is. Mistral is just the LLM, i feed it anything i want through the user prompts, it knows the objects im holding, it knows who i am. I wrote code to detect objects and faces, it knows each face in the frame and who is speaking at the time based on mouth movements. I have a DB setup so it stores facts from conversions, the conversations are far more intelligent than that of HAL (out of the box), I can train the model to role play so it pretends to be KITT via system prompt. And you can definitely give it emotions in its responses, It's all in how your prompt engineer it. Everything you listed can be done using Mistral with the help of python. So if you are saying just using the LLM out of the box has none of that, then fine, yeah it's just a text bot. But my whole gripe was that HAL is not more intelligent then these LLMs. And the fact that a dev can run a decent size LLM locally without the cloud is a game changer.
I'd love to see all your updates with HAL and please dont think im trying to put HAL down and all the hard work that has been down, i'm not. I'm really just providing my findings on new tech.
One problem with Mistral is the knowledge ended in 2022 so current events have to be trained (fined tuned) into the model. Ah and another issue with HAl for me at least, how can you solve the problem of adding custom tts voices to HAL? Right now it's only able to use SAPI voices, right? I cloned KITT's voice using coqui xtts which i can use with my bot but not in HAL.
And what do you mean by edit Mistral, its brain? why would you, you just need to keep training it. you can fine tune or train it further using LoRa. I can have mistral do anything i want it to using python, if i say "start your engine" right now, Mistral matches the command to the list of commands i have stored in a python script, then using parallel port control the car starts up, just one of many commands it will run. This is the same thing i use to do with HAL. The list goes on and on.
And when i said end of an era, i meant with my project, yes its the end of a long relationship with HAL because it just doesn't stack up intelligently to these LLM's now. I have complete control over my chatbot, its completely written in python. One thing i wanted was HAL to be ported to python so it can run seamlessly on Linux. (a ton of work im sure).
again, i'd love to see some of your updates! And cheers to all the hard work you've been doing! I love checking in here to see what kind of updates have been made.
@sybershot Thank you! it's coming along nicely. I love HAL, it was a game changer when it came out! I was using it in my car for over 10 years. I think it was on HAL 5 at the time. I had HAl controlling everything on my car from the scanner to the engine starting up. I just decided to go my own route and build my own custom bot so that i have complete control over every aspect of it.