Vinicius Alvarenga (@vinialva): "My point is that intelligence can be perceived in things that are not intelligent.
AGI might never be achieved with current architecture. Sure, depends on the definition of AGI. I envision it as remarkable as described by the AI lab CEOs… basically as a God inside a GPU.
Th…"
Home
Subscriptions
Chat
Activity
Explore
Profile<br>Create
Make money doing the work you believe in
Start your SubstackLearn more
For you
Vinicius Alvarenga Aug 11<br>@vinialva
My point is that intelligence can be perceived in things that are not intelligent.<br>AGI might never be achieved with current architecture. Sure, depends on the definition of AGI. I envision it as remarkable as described by the AI lab CEOs… basically as a God inside a GPU.<br>The 2 trillion dollar bet [so far] is that this is possible. They believe that If you have enough computing and enough parameters you can model cognition and scale it more than humans combined.<br>LLMs don’t know what they are doing. They run a predictive algorithm not a cognitive one. Kids eventually learn possibly the most important question early on. They keep asking why for everything. Their cognition isn’t after copying the most likely output. Their cognition is after a true understanding of why things are the way they are.<br>LLMs don’t care about the why of anything. All they care is that, given your token, plus their learned params, that their output is the most likely.<br>Sure the more params you have in your model, more “information” they will consider when processing your prompt. And in the middle of that , it might have stored some of the why s. But that why isn’t the major driver in its output, it is just statically there. Actually it is not even “there”, it was deconstructed into numbers / activation functions that when used helps produce answers mathematically close to something we recognise as why.<br>Can you reach AGI if you don’t understand, truly and deeply the why’re you doing anything?<br>Yann LeCun has this ideia that AI needs to understand the consequences of its actions. Which makes sense. But to me they have to understand why first.<br>To my starting point, if someone is given a Galton board and the pins are hidden from them, just input, blackbox and its output, someone would say “wow, this machine is thinking and organising things for me”.<br>The knowledge was in its construction, its opaque execution reveals intelligence, not in the in the execution, but in its construction. The output might be useful or not, but replace the blackbox with an LLM and you will get my point.
Aug 11<br>at<br>5:14 PM
Relevant people
Log in or sign up<br>Join the most interesting and insightful discussions.
Start your SubstackSign in
Get app
This site requires JavaScript to run correctly. Please turn on JavaScript or unblock scripts