The Beave's avatar
The Beave
TheBeave@GitCitadel.com
npub1q6ya...t3sh
Not much to see here... move along. #weldstr #welding #beexcellent #buildthefuture #diy #solar #alexandria #gitcitadel Solar Sovereignty group link: https://smp12.simplex.im/g#EyopPz7gbIXw2fUJLhCCfTinUpBCzZ0oyhiwF9cs_IY Catch me on simplex: http://w3.do/F9zAICbf
The Beave's avatar
The Beave 2 months ago
Warning: Ignorant ranting ahead. Proceed with caution, several grains of salt, and an understanding that the writer fully knows that he is an idiot and doesn't understand what he is writing about. I am utterly gobsmacked to learn that LLMs cannot recall any of their training data. WTF! Yes, I know this much shock you at how ignorant I can be about this, but . . . This is utterly stupid. All o fthis effort is going into creating a thing that generates . . . whatever . . . based on assumptions of relationship between snippets of truncated information. WTF is actually going on here? This is so utterly stupid I can't even . . . LLMs are worse than retarded. They are poisonously retarded. What is the point of using this technology? I don't want the world being changed by assumptions and vagueness. Why is so much being spent on technology that can't even tell you how it got to the point where you are interacting with it? The people who control what data is being used to "train" these models have to know how stupid all this is, unless they are also delusional themselves. I find the later to be more likely at this point. This isn't "knowledge." I don't have the words to describe this, currently. Matrices of truncated data are not what I want to build anything on. Displacing millions of peoples' livelihood for literal "handwavy BSery" is madness! People PAYING for their own usurpation is even more evil. (If you purchase electricity, you ARE paying for this. If you are building new hardware systems you ARE paying for this. Heck, if you are building new hardware systems with USED hardware, you are paying for this.) Data used to be getting cheaper, and now it is costing people's lives. I don't see any hope left in this most recent opening of Pandora's box. There is not much point left in traveling down the path that this leads to since it is all delusional nonsense. I cannot wait for the infinite money glitch to fail spectacularly. I am correct, in my estimation, to be moving my life in a direction that requires less computing and less electricity, and less electronic connectivity. It will certainly be less convenient in some ways, but it is certainly more robust to spend time building personal skills, systems of actual resiliency, and community.
The Beave's avatar
The Beave 2 months ago
Does anyone host their own agent stack? I've just been informed that I could possibly host an agent to interact with a vector database to solve my issue LLMs being utterly retarded in terms of context windows. I want a thing to be able to reference a massive library of works, and LLMs can't do that at all. Has anyone done this? If so, what dies the tech stack look like? (Mostly wondering about the software needed to tie it all together, hardware is in the works.) #agent #agents #selfhosting #homelab #AI #asknostr
The Beave's avatar
The Beave 2 months ago
I have found my most major frustration with LLMs: The lack of context. Small context windows and no memory beyond a few thousand tokens is utterly useless for anything I would want to use an LLM to do. Are there any models that actually have useful memory? If so, is is possible to self-host them using hardware that doesn't cost thousands of dollars?