Max Curiosity Clips
Region: AU
Saturday 30 May 2026 22:24:49 GMT
556232
19977
932
2411
Music
Download
Comments
and.eee_ :
It’s just a neural network ffs
2026-05-31 14:15:47
748
Niniki :
a processor
2026-07-30 16:19:53
0
sgaksgok :
JUST TO AVOID NEW CONSPIRACY MISTAKES… we know exactly how it is made, because we design them. What is not possible to know is the value of the weights within the cells of the numerical matrices, due to the very high number of parameters. For Gemini, for example, we are at 1.5 trillion parameters and obviously it is not possible to know exactly in which cells a certain piece of information or concept is compressed.
2026-07-30 09:11:52
1
Shrooms man! :
We don't even fully understand the Human brain lol.
2026-07-30 03:22:21
2
Typical I :
Of course we know how it works, we created it….
2026-05-31 14:51:34
1359
manhwa_verse :
neural network is actually not that deep 😑😑😑
2026-05-31 10:54:47
67
Mariia_YourAI&Security :
we do understand how it works
2026-05-31 03:10:28
215
xqzj :
they do understand exactly how it works. but they can't trace where the information lives during inference.
2026-05-31 14:12:02
175
jeronimovm :
as a tech , we DO KNOW how it works, we wrote every equation, but can't explain what they learned. The mechanics are fully transparent, forward pass is just matrix multiplications + nonlinearities, backpropagation is just the chain rule, optimizers are deterministic. Zero mystery there. What breaks interpretability (many influencers have missunderstood this and got to the erroneuos conclution that not being able to have interpretability of the weights means not understanding what is going on which is a wrong statement) is distributed representations: after training, concepts aren't stored in individual weights, they're smeared across millions of parameters interacting nonlinearly. You can't point to weight #28 or any other in the hidden layers and say what it means. Compare to decision trees or linear regression (other forms of machine learning methods that don't user neuronal networks)— those retain interpretability because the hypothesis class forces learned parameters to stay human-readable. Compare to brains or evolution — those never had interpretability to lose, we don't even have the equations. Neural nets are uniquely strange: root access to every weight, but the solution the optimizer found inside those weights is opaque. That's exactly why mechanistic interpretability research exists — reverse engineering the learned circuits post-hoc, doing neuroscience on a system where you actually have the source code.
2026-06-10 14:59:58
7
Andyy :
We greated that literally. We do know excatly how it worka
2026-05-31 08:58:15
22
evade.333 :
As someone who built and train small networks its not that we dont understand AI. We understand the calculus behind it perfectly. The issue is purely interpretability at scale. When a model makes a mistake, you can't just go in and tweak a single neuron. Instead, the industry fixes errors the practical way: by curating better datasets, retraining, and using external code as a 'gatekeeper' to filter out bad outputs.
2026-05-31 19:49:20
48
ontotheology :
calling matrix manipulation the same thing is what your brain does to form a thought it's like comparing checkers to Red Dead redemption 2. the two things are in no way analogous to each other. chat GPT and other LLMs do not have thoughts. they don't have intent, they don't have a persistent self, they don't have anything you need to have a real intelligence. That's why they measure these things in functional intelligence terms.
2026-07-11 22:43:36
10
Syndicate :
if we didnt understand how it works then how did we build it
2026-05-31 22:39:58
7
vvazrzecznyniebezpieczny :
we do. pls stop scaring ppl. its not bad bc "we dont understand it". and some instances are good. just not llm and commercial genai
2026-05-31 18:37:43
8
Joggt :
Actually we know exactly how it works. The term “black box” or “we don’t know” is just metaphorical definition since we use probabilities of millions of parameters. meaning it takes very long time to compute by hand. hope this helps kids.
2026-05-31 17:27:50
9
GuiguiLabuse :
no one in the comments, knows a damn thing about what he's talking about
2026-06-01 08:25:51
11
fartbox :
It is simply multi-dimensional curve-fitting.
2026-05-31 14:23:13
23
Lotnik :
yeah we know excatly how it works
2026-06-01 08:19:58
39
Ethan Sage :
Eso es complicado mejor hagan lo que hace la elite, es más fácil escanear cerebros 🧠 y luego pasarlos a los droides
2026-05-31 18:32:58
10
Number Black :
This is what the hard drive looked like. You don't understand exactly how it works.
2026-05-31 17:42:02
76
Seggaz :
Yeah and if we could see the raw connections of the human brain they’d be even more complex. But we’ve yet to fully map the brain at a cellular level. Roughly 86+ billion neurons and hundreds of trillions of synaptic connections. It makes our current largest AI models look like toys by comparison.
2026-05-31 08:12:30
30
Battler :
We don't understand the brain either🤷🏽♂️
2026-05-31 12:10:40
19
i bakero :
juat probabilities and testing in crazy speed a lot of electronics works like that and all just because speed of light
2026-05-31 00:35:49
10
Preston jhagroo :
Routing, mapping, inferencing, connecting. I also know nothing about this so I’m just talking to talk
2026-07-12 18:27:29
6
Mgosth :
We know EXACTLY how it works 🙂
2026-06-01 14:04:08
6
To see more videos from user @maxcuriosityclips, please go to the Tikwm
homepage.
