Am running the chinese ai deepseek-r1 on my gaming rig specs:
cpu ryzen 5 7600x3d
gpu nvidia geforce 1080 ti 8gb
ram 128 GB
am capable to generate tokens at a rate of 2-4tokens every 30 seconds or so(roughly 5 words a minute) yeah painfully slow and so far kwisha gpt everyone who works at gpt ako jobless the 70b model although not the complete model is giving me near chat gpt performance. I use gpt to summarise articles but upto now I have been using my local deepseek na iko chonjo kabisa.
I will buy a better gpu like a 4090(maybe 2 of them at the same time) and that might improve the speed.
I canāt believe one AI just sent the other AI homeā¦enyewe tunaingia era ya kupimana IQ capabilities, users washaonja and expectations are heightened.
my motherboard officially supports 196gb but i only have 4 slots so i use 4 32gb ddr5 ram sticks
Kwisha we can ran a complete model with consumer hardware, I can not remember when i last used the big techs model also qwen-30b is a beast facebooks llama-70b is good too.
lol hizi mkuu ni LLM not AI they are predictive text they can not and will not develop reasoning capabilities
In layman terms tokens is what you enter when communicating with the llms (chatgpt e.t.c)it is also the text they output example āI am a very helpful aiā contains 6 tokens āiā āamā āaā āveryā āhelpfulā āaiā
What do you mean by this? Reasoning capabilities depend on how you can process data thrown at you. For example, your reasoning capability depends on your past and present experiences. If you feed these machines with just enough data, they can āreasonā based on the pattern.
tafuta any b650m motherboard micro boards they can cost as low as 12k motherboard kama hiyo haiezi maliza wiki kwangu, futhermore most of my customers are students hizo vitu wengi hawana
Large language models are predictive algorithms and they are good at what they do, but for you to use chat gpt it has been heavily fine tuned these is what we call parameters about 8 billion for chatgpt, this is why people went crazy for deepseek it has only around 600 million parameters and offering chat gpt performance with smaller tweaks. It behaves they way it was trained to behave they lack independent unique thought. Think of it like a magic trick that you do not see how the illusion was performed.
Reasoning Ai use something call neural nets(like your neurons in the brain) and algorithms like deep learning, they study and learn by themselves without human intervention but at best the are below mediocre, the entire computing industry will need to change to create digital life am talking things like quantum computers and breakthroughs in neural net technology the computer and software that will create a real thinking ai will not look like modern day devices. software can not and will not think it just does what it was meant to do.
Do you know scientists do not know what a thought is and we have a vague idea of how it is created. is it energy? an atom? electrical impulses?
You have ai in your android phone(a very very small one) it tries to predict your text as you type based on all the data entered in you phone is it thinking???
We have had ai in airplanes that can fly the plan unassisted, self driving cars all these are ai they are just doing what they were trained on, for sure ai will change the world but they wonāt be our equals they will be good at certain tasks and thatās what they will ever be.
Niaje @fundimnoma. Iām looking for a small custom rig I can use for basic dev work kama HyperV, WSL, Docker, MS Visual Studio and such bila machine kuhema. Pia kama inaweza run small models kama Deepseek na Llama inaweza kuwa sawa. Do you sell those ama can you point me in the right direction?
You seem very optimistic that ai will not one day become self aware. As you mentioned, we donāt fully understand the concept of consciousness and we may one day intentionally/unintentionally create it
Iām no expert but the little Iāve gathered online is that the current direction weāve taken with building AI limits the development of models that would attain AGI let alone consciousness. Kuna prof flani anaitwa Yann LeCun and heās the current head of AI at meta. He drops great nuggets on this topic pale X na Bluesky.
