Thread 107154335

What's missing from the current AI and why haven't we achieved AGI?
Don't tell me it's missing sovl. I need constructive answers that can generate long debates.

>>107154335 (OP)
>give me your bilion dollar ideas
no
you give us yours and we'll call you a nigger regardless if theyre good or not

>What's missing from the current AI
everything? an llm is an interface.

>He contends that AI systems must be grounded in physical reality through sensory experience
"bro, i heard you like ml. so i put ML in your ML"
im not sure i would like this lecunn guy

>>107154335 (OP)
>faggot in picture
LLAMA 3 and LLAMA4 are trash tier models that dont produce anything.

>whats missing from the current AI?
Tiny general AI that can tool call (for extreme speed and utility), compress/quantize information in memory (for general knowledge storage), ability to search through quantized memory (for searching through memory)

>>107154335 (OP)
The von Neumann architecture is now 80 years old. It is completely unsuitable for mapping neural networks, and there are still retards who cling to it.
Find any fcking paper on neuromorphic chips, an architecture optimized for neural structures. Recognize the difference, which is as big as that between a steam engine and a jet engine.

Fcking retards, what do you do all day?

https://arxiv.org/pdf/2510.06721

>>107154467
>in order to calculate how long it will take for a car to go from point a to b, we will now proceed to build a car and drive it from point a to b
nigger

>>107154490
> An infinite number of monkeys typing on an infinite number of typewriters would almost surely type the complete works of Shakespeare.

keep scaling, eventually it'll work out with the monkeys

>>107154547
what if all the monkeys do is iterate between the letters e, g, i, n and r?
kinda like cyclical numbers
or even random ones but with missing keys and no way of knowing that?

>>107154335 (OP)
I dunno though. I had a good experience with Gemini the other day. The problem is when LLMs are framed as super intelligence. They're not. But as a natural language search engine they're ok. They make shit up a fair bit, but so do people. Just double check what they output. It would be nice if their developers didn't try to make them appear as the font of all knowledge however. I can read the first paragraph of wikipedia articles for myself.

Back in the day, telephone operators didn't claim to know everything. LLM assistants should be like them.

>>107154547
>>107154565
forgot picrel-schematic of the idea

>>107154335 (OP)
The guy you posted has the answer, OP.

https://www.youtube.com/watch?v=yUmDRxV0krg

>>107154575
>use a statistical model to compute deterministoic phenomena
hes a nigger. thats toroughly retarded

>>107154335 (OP)
It's a dead end, but we don't seem to have reached the end yet.

AI needs to:
-Learn at all times - it's ridiculous to have training and deployment phases, it should be able to adapt to situations that were not in the training set.
-Be able to observe the results of its actions more directly, through a sensory system, that is something more than an idiot typing in vague prompts
-Have goals which are something more than : "produce text that is most likely to make carbon based life form happy"

>>107154598
and then people pretend like big corpos management isnt completely fucktarded
using big corpos as example to follow should be a bannable offence

>>107154638
its already more than that, on all accounts,
nigger aicultist rabble

>>107154640
Come on. Just buy more of Jensen's GPUs. You don't want to be left behind, do you. Everyone's pumping billions into AI. Your stock price will fall unless you can show you're utilizing AI.

>>107154652
but we do use ai.
and thanks to our innovative marketing department, you can too
for the low price of your left ball and your eldest son's soul!!!

>>107154638
Humans have two goals (or prompts): survive and reproduce. All of our behavior and everything we recognize as intelligence emerges from this.
The basis for AGI must be radically simple; if nothing emerges from it, we are on the wrong track.

>>107154676
I genuinely can't tell if this is a vaccuum cleaner accessory or a sex toy

>>107154698
>Humans have two goals (or prompts): survive and reproduce. All of our behavior and everything we recognize as intelligence emerges from this.

no, thats animals
and if you didnt grow out of it youre not human.
and also a nigger

>>107154705
its not a bug its a feature
a feature derived from leaving design to women and homosexuals

>>107154707
It's the 21st century, I've overcome my ‘god complex’. I can accept myself as I am.

>>107154728
no, its just cope for not being strong enough to strive for more
case in point:
has fuck all to do with god complex
you cant even be fucked to look up the terms youre using

>>107154490
>in order to calculate how long it will take for a car to go from point a to b, we will now proceed to build a car and drive it from point a to b
>nigger
No, anon was talking about building/driving a car not stealing a bicycle

>>107154792
niggers come in all shapes and colors

>>107154638
The entire thing is out of wack. If LLMs were the future, why can humans learn a skill with far less input, using far less power during training/inference than the token predictors?

>>107154841
our whole brain runs on less energy than a lightbulb
that argument can be waived away with the higher efficiency of biological computers

>>107154841
>>107154862
cont.
there is a reason for that
one thats known by some elements in the industry
im not getting paid to give ideas though
so ill just shoot down the bad ones and call people various ethnic and or homophobic slurs

>>107154911
actually fuck the industry
ill deliberately muddy the waters
they should have lurked more

>>107154862
If that’s true, then investing in LLMs at the current hardware stage is pointless, since it implies there are ~10^10 improvements to the hardware side.

>>107154821
can't get a nigger in white though
inb4 michael jackson

>>107154990
and thats the reason why techbros want to go nuclear
the magical thing about bruteforce, is when it is applicable, the problem can be solved with enough of it
problem is that theyre investing hundreds of billions into what amounts to an evolution of the keyboard

>>107155043
you sure about that?
>Historical Genetics: The Genetics of the Jews. A. E. Mourant

>>107154335 (OP)
Its very simple, LLMs construct and replicate patterns at the syntactic level, not the conceptual level. It can tell you "paris is the capital of france" not because it has a model of nations, capitals, and a fact set that associates those two places with that model, but because words like that appeared in the training data. It can tell you that simply because it was told so.

It is the very definition of a midwit: Smart enough to tell you everything its been told, not smart enough to understand any of it critically, much less do its own research or learning.
It is quite poetic that reddit makes up such a large part of modern models training sets, because you could not build a more accurate parody of a redditor in machine form.

>>107155348
People buy things they don't need to impress people they don't like.
And the only reason for this is that they have been conditioned to do so. The human brain conditions itself to inputs from the environment. So a world was created that provides such inputs.
Then there are retards who think they are free and spend half their day thinking about how to express their uniqueness by buying what a bunch of wankers have decided they should buy, along with millions of others.

>>107155496
>>you can go above and beyond
>no, because theres people who dont
ok, dont then. but you wont get me to say "yeah, thats unavoidable" bc it isnt.

what youre describing is an immature ego where the sufferer relies on the way hes perceived by others to attribute worth, or lack thereof to himself
soiciety promotes this mental illness because that makes good consumers