@alberta.nyc: No, building a chatGPT wrapper does not make you qualified to tell me how LLMs work 😌 The full vid is in my bio #ai #emergentcapabilities

Alberta Tech
Alberta Tech
Open In TikTok:
Region: US
Sunday 08 June 2025 23:59:03 GMT
79499
7999
321
202

Music

Download

Comments

sheebdog
sheebdog :
I mean tbf, he’s a ceo, not a technical lead…
2025-06-09 00:26:22
685
natenorthway
Nate Northway :
I understand how it works (it doesn’t) hope that helps
2025-06-09 00:04:54
763
jothebuzzyard
JustJo 🇸🇪🏳️‍🌈 :
I mean I don't find it surprising the CEO doesn't work but a random redditor does. CEOs rarely know how anything works
2025-06-09 05:56:14
139
masterofspifness
masterofspifness :
Many people refer to AI hallucinations as if they're bugs. No. The AI is ALWAYS hallucinating. When it happens to produce something useful, either there are checks to prevent obviously useless stuff from being sent, or it hallucinated something that luckily was useful.
2025-06-09 03:49:06
95
appleuser125389493926627
Shiro :
Transformers are open source and free, the only proprietary piece is the trained weights. Once you scale those weights into a massive, high-dimensional space, you get emergent properties a phase transition surprise where the model suddenly writes essays or jokes you never explicitly taught it. And newsflash: we fully understand how that happens it’s just large scale pattern learning in math. It’s basically the same phenomenon that makes humans learn language and concepts, only here it’s silicon and tensors instead of neurons and synapses.
2025-06-09 09:23:24
12
turingcomplete19
turing-complete :
they are not pre programmed, true. they are emergent properties in complex systems. but they still work by virtue of the math behind statistics, SGD and everything else - which is very well known and understood.
2025-06-09 06:08:19
57
meanylinguini
meanylinguini :
I'm autistic. When I interact with AI it's clear that it's attempting all possible methods and using what works, similar to autistic thinking.
2025-06-09 01:09:47
6
moteofdustinasunbeam
N :
3blue1brown knows
2025-06-11 14:47:08
3
brucephalus
Bruce K Logan | 🧝‍♂️🧙‍♂️ :
Eh. Most CEOs don't know anything about their company's products.
2025-06-09 03:16:50
11
diegomolina18980
diegomolina18980 :
It makes sense that the CEO does not really know how it works, right? How many CEO’s understand how an API works? Btw, AI is just pattern recognition
2025-06-09 15:51:28
4
omgerez
Erez :
Altman said the same, that in the end it's a lot of magic they do not fully understand
2025-06-10 05:45:47
3
noelle_utdr
Noelle :
professor aftercare killed me 😭
2025-06-09 07:24:52
27
prof_info
prof_info :
In the professor's defense, the person also understands string theory, time, income tax and rocket surgery.
2025-06-09 18:30:20
2
justhereforthedulcimer
JustHereForTheDulcimer :
Ai is not auditable or traceable - nobody knows how it works directly
2025-06-11 19:39:08
2
poshj123
JP :
it's just predictive text. I've been using it since T9 flip phones.
2025-06-09 17:46:32
4
iseelivingsouls
MrLeast :
The output isn’t deterministic is what it means. It can’t be otherwise it’s a search engine.
2025-06-09 21:02:44
2
gimpytheterrible
Gimpy The Terrible :
Apple seems to think it is just really really good at pattern matching, not “reasoning” 🤷‍♂️
2025-06-10 02:54:30
3
drprofessorcath
DrProfessorCath :
My heart dropped when you said “random redditor named professor” I thought someone found a deep comment of mine! But no I also don’t know how they work and I’m a GenAI data scientist.
2025-06-09 11:02:00
5
mlogan175
mlogan175 :
Sounds like they are confusing "I know how to work with AI" and "I understand why these billions of parameters work and why we get the results we get"
2025-06-09 02:34:25
8
lucys372
lucys372 :
Reddit in a nutshell
2025-06-09 01:34:00
6
chizzles09
Charlie 🍉 📼 :
we understand the training methods, we do not understand the outcome of those training methods, just that they generally predict the best/most likely outcome.
2025-06-09 06:07:43
6
akumpo8ft
r :
this is a nuanced topic. on one hand, developers obviously know the architecture of LLMs and the reasoning behind why they are designed that way. but it also gets to a point where one can only make an "educated guess" on why it works, based on empirical evidence. you can never truly get a full understanding of it, especially as it gets larger and larger.
2025-06-10 03:18:48
7
johncolins41
johny :
we did train him for what you have said. we know why it works but we definitely don't know how lol. understanding a neural network is like a train to understand some people's brains by holding it. basically impossible.
2025-06-09 00:53:05
5
lurking_leviathan
Lurking Leviathan :
In all honesty, I'll believe we don't know how it works when Greg, the guy who spent 400 hours building it tells me that. A CEO has the credibility of a cardboard box.
2025-06-09 20:05:14
1
waterursa
Waterursa :
I understand it but I don’t feel like telling you
2025-06-10 06:26:02
1
To see more videos from user @alberta.nyc, please go to the Tikwm homepage.

Other Videos


About