The 90-year-old idea behind JEPA models: Canonical Correlation Analysis
https://shonczinner.github.io/posts/embedding-prediction/
#HackerNews #JEPA #models #Canonical #Correlation #Analysis #data #science #machine #learning #research
The 90-year-old idea behind JEPA models: Canonical Correlation Analysis
https://shonczinner.github.io/posts/embedding-prediction/
#HackerNews #JEPA #models #Canonical #Correlation #Analysis #data #science #machine #learning #research
Here we have a classic example of the difference between causation and correlation.
This article is titled: Could the iPhone be to blame for America's plunging birth rate?
Ummm, no.
You know what else happened in 2007, right? The economic crash that saw millions of people lose their homes, their cars, their good paying jobs, and their future, essentially.
Of course they don't want kids. The economy has never really recovered.
Link: https://sherwood.news/world/could-the-iphone-be-to-blame-for-americas-plunging-birth-rate/
It is easy to stereotype, but these some make it easy
I am sure people will be reassured that 4 out of 5 had not been reported for domestic abuse. And while reported does not mean convicted, not being reported is not a guarantee that they have not been yet.
Lots of nots there but there is a strong correlation between domestic violence and rioting neds.
If scale blindness explains why humans mistake local patterns for universal laws, correlation‑causation confusion explains why humans mistake coincidence for intention. #coincidence #correlation
https://survivorliteracy.com/2026/04/24/chapter-8-correlation-causation-and-confirmation-bias/
This article explores the 3 vital rules of science. Together, they offer a powerful lens for evaluating the validity of claims and conclusions.
#Science #CriticalThinking #Falsifiability #Replicability #Correlation #Causation
Ultimately, any "intelligence" we perceive in these models is projected by us. The model only understands the mathematical links between variables and outcomes. As George Box famously noted: "All models are wrong, but some are useful."
This tread is about generative AI and AGI in general and Large Language Models in particular.
#machinelearning #ai #llm #correlation #causalinference #causation
5/5