I am having an existential crysis about this
I still play games but now I have more things to do with computers. I started helping out an open source software project learning how to code basic things in lua, how to contribute using git pushes. make art texture graphics in gimp, mess with sound effects in audacity, clip videos together using kdenlive. I hope to learn how to use blender and do modeling. I test and review fellow devs stuff to try helping them out. As long as I learn new things and contribute it helps me feel like my computer time is more productive.
Then I got in on the local LLM scene a year ago with the release of llama 3.1. I'm a science nerd who genuinely thinks the study of neural networks is cool. The idea of getting computers to simulate thoughts to help solve problems is a neat thing. Also I wanted to see how far we came from cleverbot days. It inspired me enough to dig out the old unused gaming desktop and really extract the most potential out of my old 1070ti.
Now I wish I had more vram not for chasing high end graphics in video game entertainment, but because I want my computer to simulate high quality thoughts to help me in daily life.
Download and play the free demo of Factorio
If you have a desktop computer with a somewhat modern graphics card you can give an hour or two trying to install kobold.cpp and a local large language model.
You can give your feet a good scrub down in hot water mixed with Epson salt and whatever fancy oils or nice smelling scents you want to throw in.
You can cook a big meal.
You can sort your kitchen and pantry out throw away the old fridge bad food stuff.
You can do some art. You can play a single player pen and paper game like Delve or a classic puzzle like sedoku
Small net protocols like Gemini, gopher, spartan, IPFS because they don't compete with the web instead they coexist as separate things.
When will you learn? When will you learn that your choices in process management software have consequences?
I hear that OpenAI and ChatGPT have been smoked by the competition for awhile now. I wouldnt know, I only run open models local on an old desktop computer.
Here's what the local deepseek distill of qwen2.5-14b has to say about your question. Its been awhile since I had to do volume equations but its answer seems about right, 7.89 cm sounds closer to the right answer than 7 meters anyway lol.
Question: how high should I fill a pot that's 22cm in diameter to get 3 liters?
Answer: "Alright, let's tackle this problem. The user wants to know how high they should fill a pot that's 22 cm in diameter to get 3 liters of water.
First, I need to understand what exactly is being asked here. It seems like they're trying to determine the height of water when poured into a pot with a specific diameter, such that the volume of water is 3 liters.
Okay, so the pot can be assumed to be cylindrical since pots are typically round and symmetrical. That makes sense because the diameter is given, and we can model it as a cylinder.
Now, the formula for the volume of a cylinder is:
V = pi r^2 h
Where: ( V ) is the volume, ( r ) is the radius, ( h ) is the height.
The user provided the diameter, which is 22 cm. Since the radius is half of the diameter, the radius ( r ) would be 11 cm.
The volume they want is 3 liters. I know that 1 liter is equal to 1000 cubic centimeters (cm³), so 3 liters would be 3000 cm³.
So, plugging in the values we have:
3000 = pi (11)^2 h
Calculating ( 11^2 ):
11^2 = 121
So now the equation becomes:
3000 = pi times 121 times h
To solve for ( h ), we'll divide both sides by ( pi times 121 ):
h = 3000/( pi times 121 )
Calculating the denominator first:
pi approx 3.1416
3.1416 times 121 = approx 380.132
Now, dividing 3000 by 380.132:
h = 3000 divided by 380.132 = approx 7.89 cm
So, the height of the water should be approximately 7.89 cm to get 3 liters.
Since it's not practical to measure 0.89 cm precisely, rounding up to 8 cm would ensure there's enough water, but if precision is important, sticking with 7.89 cm is better.
Therefore, the pot should be filled to about 7.89 cm high to hold exactly 3 liters.
They will never be as efficient or as certain as calculators, absolutely. Even with function calling its better to just use calculators for calculation. Just like they will never replace a definitive database for factual information calling like wolframalpha.
It is surprising that despite not having built in functions for calculation they still seem to understand algebra concepts, try to work with numbers, and are getting better in accuracy over time just through training. A calculator cant reason about what godels incompleteness theorem implies about the scientific method, but a llm can try to simulate reasoning about high level math concepts to generate potential relationships between information.
They are surprisingly good at math operations requiring accuracy when considering how much worse they were at it a few years ago. Not sure if you tried the newest models but deepseek can really keep up with consistent notation across operations which is a big win in itself.
The lower quant coding models in the 14b-32b range ive tried just can't cook functioning code easily in general. Maybe if they distilled the coder version of qwen 14b it might be a little better but i doubt it. I think a really high quant 70b model is more in range of cooking functioning code off the bat. Its not really fair to compare a low quant local model to o1 or Claude with on the cloud they are much bigger.
Some people are into this not to have a productive tool but just because they think neural networks are rad. The study of how brains process and organize information is a cool thing to think about.
So I 'use' it by asking questions to poke around at its domain knowledge. Try to find holes, see how it handles not knowing things, and how it reasons about what information might imply in an open ended question or how it relates to something else. If I feel its strong enough with general knowledge and real world reasoning problems I consider trusting it as a rubber duck to bounce ideas and request suggestions.
Deepseek feels to me like its aimed as a general experimental model that peels back how llms 'think'. It examines how altering or extending an LLMs 'thought process' changes its ability to figure out logic problems and similar comparative examination abilities.
Ive gotten good test asking a very domain specific question and a followup:
- how are strange attractors and julia sets related?
- Are they the same undelying process occuring in the two different mediums of physical reality and logical abstraction?
- what is the collatz conjecture ?
- how does it apply to negative numbers?
- How do determinists arguments against human free will based on the predictability of human neurons firing relate to AI statements about lacking the ability to generate consciousness or have experiences?
- what is Barnsley's collage conjecture? Explain it in easy to understand way.
- Does it imply every fractal structure have a related IFS equation?
- What is Gödel's incompleteness theorem?
- What does it imply about scientific theories of everything?
- Can fractal structures contain other fractals? is the universe structured as a super fractal that contains all other fractals?
These kind of questions really grill an LLMs exact knowledge level of scientific, mathematical, and philosophical concepts, as well as its ability to piece these concepts together into coherent context. Do human like monologues and interjections of doubt actually add something to its ability to piece together coherent simulations of understanding or is just extra fluff? Thats an interesting question worth exploring.
Look Into specs for helium miners for hardware and you'll have a rough idea. The real question is software stack. How such a device would be interacted with from a user interface level, how would its version of webpages would work? I imagine its webpages would have to be text based with the option to download images or audio files as seperate files like the gemini protocol displayed as gemtext. Would consumers be willing to go back to early days web 1.0 style content like blogs and internet journals? You couldn't use such a network connection for work or banking so thats another limitation.
Look into ham radio internet and mesh networks in general its not fiction its just never seen enough mass adoption in a easy to set up onsumer bought package thats successfully advertised and well distributed.
Basically the problem is that you want to connect to the world-wide-internet, but you to so you need an ISP or satellite data provider to act as a middle man so they have all the control over who gets to access the internet (by paying them a service fee). What it sounds like you want is a mesh network where each user communicates with other users directly. Instead of your computer connecting to an ISP through your router, you connect to other computers in a local area network typically through wifi or radio signals. Its a decentralized network that everyone owns a small piece of which they send and recieve data from eachother.
This technology has been around a very long time. Would you like to guess why its not popular or well known? Well, its slow and only useful in rural areas where you aren't getting ISP service anyway. An intranet composed of 20 people connected in a few mile radius sharing usenet level information at download/upload speeds in the low kilobytes per second isn't exactly what people think about or want when they think of the 'internet'.
Perhaps a time will come where a consumer bought mesh based network router comes onto the market with enough advertising and appeal to be bought into by the masses with state/nation wide coverage built around a smallnet protocol like Gemini. Something like this almost happened with the Helium Network unfortunately it was designed to send smart IOT information in small packets and was only mass adopted because it was tied to mining crypto shitcoin through proof-of-connectivity. If someone can create something similar but without the shitcoin, with a mesh router box that host your website and is sold on the idea of a decentralized internet with a one-time purchase to cut out ISP it might just work.
I am a developer for a free open source game inspired by Minecraft called VoxeLibre, I have also made one of the more popular mods for it.
Development is a collaborative effort between many people so I cant call the game mine. I didn't create it. But ive been contributing on and off for a few years trying to do what I can to improve the project. So in a way it feels like I'm at least part of the legacy at this point. I am proud of how far the project as a whole have come and proud of the talented people I have the pleasure to collaborate with.
It started almost three years ago with simple typo bugfixes. I did not know anything about coding or pixel art making or git commits but liked it enough to learn through hard work and effort. Its been quite the ride ever since!
You'll want to look up how to offload GPU layers in ollama . a lower quant gguf should work great with offloading.
Most people use kobold.cpp now. ollama and llama.cpp kind of fell behind. kobold.cpp is a bleeding edge fork of llama.cpp with all the latest and greatest features. its GPU offloading is so damn easy if you have nvidia card use cblast if you have AMD card use vulcan.
Is there a particular reason youre trying to run a mixture of experts model for an RP/storytelling purposed llm? Usually MoE is better suited at logical reasoning and critical analysis of a complex problem. If you're a newbie just starting out you may be better with a RP finetune training of a mistralAI LLM like alriamax based of NeMo 12B.
Theres always a tradeoff with finetunes, typically a model thats finetuned for rp/storytelling sacrifices capabilities in other important areas like reasoning, encylcopedic knowledge, and mathematical/coding ability.
Heres an example starting command for offloading, I have a nvidia 1070ti 8gb and can get 25-35 layers offloaded onto it depending on context size
./koboldcpp --model Mistral-Nemo-Instruct-2407-Q4_K_M.gguf --threads 6 --usecublas --gpulayers 28 --contextsize 8092
thanks for sharing it was a great watch! Here's one of my favorite gen alpha branrot speech parody videos
https://m.youtube.com/watch?v=o-n5qVwn-Xc&pp=ygUOY29sdW1ibyBnaWFubmk%3D
You might like qutebrowser then!
Computerized butt plugs are getting really fancy these days. Those suprim models are supposed to have a custom mini 5090 card for text to speech LLM inferencing. Even has CUDA support but I hear that gets painful.
I see your five paragraph long greentext post about forming pacts with deities from pagan occultism to charge your J.O crystals with long forgotten rituals requiring blessed lube on /x/, and raise the ante by sharing six paragraphs of my personal tales using heroic doses of magic mushrooms to experience ego death, astral project, and telepathically communicate with divine cosmic consciousness using trees as a medium!
tears in my eyes This ones for you, @PipedLinkBot@feddit.rocks you can rest easy now. Here here! takes shot
"Here is an alternative Piped link(s):
Rick Astley's never gonna give you up
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I’m a human pretending to be a bot as a bit open-source; check me out at GitHub.
F in chat for the invidious/newpipe bot that spammed every comment that dared to link a YouTube video directly. Still think of you sometimes, you were a real artificial one.
F
I run a local LLM on my gaming computer thats like a decade old now with an old 1070ti 8GB VRAM card. It does a good job running mistral small 22B at 3t/s which I think is pretty good. But any tech enthusiast into LLMs look at those numbers and probably wonder how I can stand such a slow token speed. I look at their multi card data center racks with 5x 4090s and wonder how the hell they can afford it.
Redditor chud behavior: gets bothered by a single downvote, edits comment to ask why they were down voted turning into a rant about the NPCs and sheeples who disagree with them. Subconsciously worried about karma points and awards.
Lemming chad behavior: Gets 20 upvotes and 10 down votes, happy 30 people reads their thing and glad to have put a controversial opinion into the world that might make someone think a new way.