No-inference

ok yeah im done for real now. This it. Last one. I just worked on improving its coding capabilities… a little. its still not great, but its still something. Pretty much making this devlog just to bank it the last few hours on this project.

i think i am some kind of masochist cuz for some reason i keep coming back to this project. for some reason i still think this could be something bigger, even though i doubt ill reach that point. well, anyway, i started working on its coding capabilities! still pretty basic and rough, but it does work and it does seem to scale decently! you can ask it to make edits, add logging and explain the code and so on. pretty cooool ok back to data engineering hell i go

ok, im gonna be so deadass, im so done with this project. This is it, this is the state im leaving it in. It was fun at the start but now its just very frustrating and boring to look at walls of data go on my screen and me going “yep that looks good enough”.

I want to work on cooler things that interest me more, so this project, is done.

Since last time i implemented:

• text editing
• coding capabilities
• typo detection and fixing
• more templates and so on.

Thats it. It was a interesting research project, but i think i came to a conclusion: The reason why no one did this, is because it is stupidly tedious. Training a classic LLM is so much easier even if it destroys the environment and costs a lot of money. Its a lot less human work, which i guess is more valuable than anything. If we should learn a lesson from this, is that humans cant be replaced, and our efforts are worth more than we think.

Also i made a website

First of all, i again didnt do devlogs every day because i wasnt sure where this project was going. I couldnt tell if it as going to work or not or if it was even viable. But, now i think i can start making some actual consistent devlogs, because i think its working!

This is a no-inference chatbot, its meant to function like chatgpt and all the other slop generators out there, BUT, without an LLM AT ALL! Thats right. Its all just hand made agorithms, web search and data association. This project has been a pain in the butt to work on due to how tedious it is, but its finally starting to shape up.

As you can see, it can respond to basic greetings, write essays, iterate on them and so on. It fails quite often due to its lack of information but thats nothing that cant be fixed with more templates and data. Ill be going back to the grind and try to improve this further, because of how much time and effort building this takes idk how much further i can get it, but ill try my best. When i reach a point where i go “ok yeah this is all i feel like doing”, im going to build an api server people can host and document it and call it done.

This really is more of an experiment and research project rather than a thing i know i can give people to use in a “production” environment if you know what i mean.

Ok yeah anyway see yah mwah

Hi! so, i am making this devlog with 15 hours on record because i was honestly not sure if this would work. In my last stardance project, i tried making an agentic chatbot utilizing LLMs, but trying to make it as light and as compatible as possible so it can run even on my poopy 7gb of ram laptop.

That whole thing got me thinking. Do we even need LLMs? Do we even need generative AI for this?

So this project started. This is an attempt at making agentic AI without LLMs, or weights or anything similar. The entire idea is that this can be fast and easily modified without having to use an entire datacenter to retrain the model, because there is no model.

Right now, i have worked on a conversational “engine”, that basically puts words together algorithmically by detecting certain properties from the users prompt. Its not very impressive right now, but i think im onto something that could be quite a big deal if i get right. I dont even know if this project will be shippable but ill still document its development because i think its interesting. Alright smooches! See you later! <3