AI for solving any problems // will machines surpass human intelligence?)
If you have a screwdriver in your hand, any problem looks like a screw
AI dystopia == “AI will affect every aspect of human activity”
Storing Energy is a big challenge, if we can do this we just can accumulate solar energy, that’s all. But how could AI help?
SSL — self supervised learning == flling the blanks
NLLB == no language left behind
Content moderation // Is AI a solution?
Image generation // that’s what AI really can do..
Predict last word // LLMs
LLMs have no common sense..
Hallucinations
Meta AI // chat bot
AR-LLMs suck! // don’t reason, plan, have no common sense
Human knowledge is not a text, but an everyday world..
It’s an open question of how the knowledge actually encoded.. text is just a one semi-structured form..
It’s not a task for an LLM to plan or reason. LLM just processes texts (from input prompt to output).
Same thing we can talk about calculators, they don’t know how the world works, they just calc numbers and don’t know the meaning of that numbers #lol.
“There is no question that, eventually, machines will surpass human intelligence in all domains..” — for sure calculators calculate numbers faster, LLMs generate texts..
Every professional will promote what he does, no surprise..
Object driven AI // actually data-driven
The success of modern computers, Internet and AI is depends on data. No matter what model you have. In any case it must be changed over time, it’s an evolution.
// if(){}else{} // IF THEN ELSE
World model // great idea, but how
We are missing something big // for sure
Moravec’s paradox is the observation in artificial intelligence and robotics that, contrary to traditional assumptions, reasoning requires very little computation, but sensorimotor and perception skills require enormous computational resources. The principle was articulated by Hans Moravec, Rodney Brooks, Marvin Minsky and others in the 1980s. Moravec wrote in 1988, “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”.[1]
MLLs, SSL, generative etc. suck.. // predicts all details including irrelevant ones (fixed)
Idea: predict abstract representation..
Play with abstract representations // predict something abstract way
JEPA // predictive architecture
SSL + JEPA // embeds context
LLMs that can reason, plan… // calculators that can..
