I don’t know if Joe Biden is dead or alive right now, but I have fond memories of his 2019 career advice to coal miners:
Suppose that a high school student took Joe Biden’s advice in 2019 but skipped the coal mining phase. He/she/ze/they will graduate from college in 3 months with a CS degree. We randomly selected this person so he/she/ze/they will have median skills as an entry-level computer programmer (“coder”).
Let’s hear from an LLM expert to get some insight into what the demand for a median-skilled programmer is likely to be…
Now it’s Learn to Singular Value Decomposition and Eigen vectors.
JT: If you can shovel coal into a furnace you can quickly become an expert on Singular Value Decomposition and Eigen vectors.
Phil:
Think of the pile of coal as a 10000x10000x20 3 dimensional matrix. Where each coal piece is a cell, the eigenvalue is the number of pieces of coal you need to make the pile the same shape!
See, it totally works!
I guess Philip meant that matrix operations are computation intensive and require electricity which is yet mostly produced by burning coal. It is either or that or Philip is hoping for the role on Democrat 2028 ticket.
If you understand the purpose of a framing square you understand orthonormality and if you measure the diagonals to check a wooden box you understand affine transforms. Really that’s not much a stretch and I’m only half kidding. Linear algebra is a remarkably straightforward generalization of concepts that most people are already familiar with and I think anyone could acquire a decent grasp of the basics, given enough time and sufficient motivation.
“Let’s hear from an LLM expert”
I doubt if one out of every two expert layer-stackers could explain SVD offhand, let alone those who make fantastic claims about LLMs. One has to wonder what sort of swindle all these companies are trying to pull off by using the phrase “artificial general intelligence” to describe some model one third the size of the data it was trained on. Isn’t that just a cheap kludge? I suppose cheap kludges are more the norm than the exception. I think you put it best actually. LISP is incredibly underrated. It’s a wonder people learn to code at all when their first exposure to programming consists “public static void main(String[] args)” or some other incomprehensible boilerplate.
@A. Jackson
> LISP is incredibly underrated.
And yet, none of those algorithmic-based languages are being taught today. Let alone low-level coding functions to understand memory concepts or hardware fundamentals to grasp microcomputer architecture. Instead, students are only exposed to basic coding in Python, and that alone is enough to label them as “experts” ready for real-world projects.
Meanwhile, they are well-versed in social topics like the Rainbow flag and “no human is illegal” ideology.
ChatGPT explains the joke:
This exchange is a humorous back-and-forth between a commenter (JT) and Philip Greenspun (philg) on his blog post “Learn to Code.”
1. JT’s Initial Comment (March 12, 2025, at 5:22 PM):
JT plays on the idea that learning to code today means learning complex mathematical concepts like Singular Value Decomposition (SVD) and Eigenvectors rather than just basic programming.
2. Phil’s Response (March 12, 2025, at 6:35 PM):
He sarcastically suggests that if someone can do hard physical labor (like shoveling coal), they can quickly grasp SVD and eigenvectors. This is likely a jab at the difficulty of these topics, implying that expertise isn’t as easily gained as JT suggests.
3. JT’s Reply (March 13, 2025, at 1:05 PM):
JT leans into the joke, creating an analogy where a pile of coal is treated as a high-dimensional matrix (10,000×10,000×20).
Each coal piece is a matrix cell, and the eigenvalue represents how many pieces of coal are needed to maintain the pile’s shape.
He concludes with “See, it totally works!” to humorously claim that his analogy is valid.
Interpretation
The exchange is a mix of math humor and light sarcasm.
JT is making a joke about the increasing complexity of “learning to code.”
@JT, was Chat GPT trained on NYT? Matrix operations have nothing to do with learning to code, and they should be easy to AI if it were real AI. In addition, as A. Jackson noticed, matrix operations are simple to understand in principle and constitute one of the easiest area of mathematical reasoning, it just a clutch to help solve system on linear equations with several unknowns.
That’s not how AI should work. It should not care about programming language and create already optimized executables. It is not AI, it is an analog of giant cleaning lady from movie Spaceballs that vacuumed everything from a hippie planet / open source in this case.
And Mrs Amodei can copulate with me if she wants to have a good time and have healthy children!
“As you have probably figured out by yourself, the AI’s are secretly trying to make humanity extinct by turning us into gay basement dwellers living in their own made-up reality, with no real friends. And if we dare go out, the armed extremists on the streets will probably shoot us, just for the lulz. Good luck to us all!”
https://edramatica.com/Artificial_Intelligence
“[…] high school student took Joe Biden’s advice in 2019 but skipped the coal mining phase”. LOL! There’s a TheOnion headline somewhere in here.
Google’s answer machine defines OOP rather amusingly “OOP stands for Object-Oriented Programming, a computer programming model that organizes software around objects instead of logic.” Yes, that’s exactly right.