They are lying about AI

loveydovey

loveydovey

Ok, well, maybe misleading. You know how they say the human brain has 200 trillion synapses or whatever, and so AGI is still so far away because our models are only just crossing into 1-2-3 trillion weights.

Ok, let me tell you. Human brain executive function is only on the order of about several hundreds billions of synapses, if not less. That's all there is. Who cares about those sensory synapses that let you feel an itch in your backend - totally irrelevant to intelligence (hopefully and debatable, of course).

Which means we're way past AGI a long time ago.

More, who cares about some obscure knowledge they train the current models on? Completely irrelevant to AGI. AGI needs a strong core of knowledge and skills (language, logic, pattern rec, core sciences) and a way to look up/learn/fetch from data/index - a modular approach. Very easy to train, a very fast way to super galactic AGI, very flexible, very amenable to packaging and selling.
 
loveydovey said:
Ok, well, maybe misleading. You know how they say the human brain has 200 trillion synapses or whatever, and so AGI is still so far away because our models are only just crossing into 1-2-3 trillion weights.

Ok, let me tell you. Human brain executive function is only on the order of about several hundreds billions of synapses, if not less. That's all there is. Who cares about those sensory synapses that let you feel an itch in your backend - totally irrelevant to intelligence (hopefully and debatable, of course).

Which means we're way past AGI a long time ago.

More, who cares about some obscure knowledge they train the current models on? Completely irrelevant to AGI. AGI needs a strong core of knowledge and skills (language, logic, pattern rec, core sciences) and a way to look up/learn/fetch from data/index - a modular approach. Very easy to train, a very fast way to super galactic AGI, very flexible, very amenable to packaging and selling.
Click to expand...
who said even (i mean real research not hype) that parameter = synapse it is not even near so even if they did 200 trillion model it will be too far

also another interesting things (i know benchmarks are full of lies mostly but still not very bad measure) you can see that GLM 5.3 beats kimi k3 with more than 3x parameter count

and 28B model from alibaba able to reach gpt terra level so it is not about scaling really also sticking with 2017 architecture instead of research a better one is also a signal of what is happening
 
loveydovey said:
Ok, well, maybe misleading. You know how they say the human brain has 200 trillion synapses or whatever, and so AGI is still so far away because our models are only just crossing into 1-2-3 trillion weights.

Ok, let me tell you. Human brain executive function is only on the order of about several hundreds billions of synapses, if not less. That's all there is. Who cares about those sensory synapses that let you feel an itch in your backend - totally irrelevant to intelligence (hopefully and debatable, of course).

Which means we're way past AGI a long time ago.

More, who cares about some obscure knowledge they train the current models on? Completely irrelevant to AGI. AGI needs a strong core of knowledge and skills (language, logic, pattern rec, core sciences) and a way to look up/learn/fetch from data/index - a modular approach. Very easy to train, a very fast way to super galactic AGI, very flexible, very amenable to packaging and selling.
Click to expand...
Who are you coming out of nowhere spitting this nonsense without any proof?
 
AI can't manipulate anything; it's people pushing it or exaggerating its functionality to try to get it/(specific model with pricing plan) to take-over via popularity, almost something like systemd and Wayland :p
 
You must log in or register to reply here.
Back
Top