The new semester started and this year I have added nine slides for my intro lecture in C++ about the use of LLMs when trying to get better at programming.
-
None of the skills you have were acquired without having to practice
You learn best when you fail.
You learn even better when you teach.
Maybe you want to become somebody who does more than just copypasting what the machine tells them to write?
2/9
We don't write software to create software in the same way like we don't cook so that there is food. Otherwise, why don't you just always buy the food
mass-produced in a factory somewhere?LLMs can generate a million paintings that look like they are created by an expert artist. Would you visit a gallery with these paintings? Why not?
A robot can outperform any athlete. Would you rather watch the human do sports or the robot do sports? Why?
3/9
-
We don't write software to create software in the same way like we don't cook so that there is food. Otherwise, why don't you just always buy the food
mass-produced in a factory somewhere?LLMs can generate a million paintings that look like they are created by an expert artist. Would you visit a gallery with these paintings? Why not?
A robot can outperform any athlete. Would you rather watch the human do sports or the robot do sports? Why?
3/9
Maybe you want your program to be more than something like a cheap T-shirt produced by underpaid labor and discarded the next day? Maybe you want to create something that lasts, which can be maintained, is secure and reliable.
3.5/9
-
We don't write software to create software in the same way like we don't cook so that there is food. Otherwise, why don't you just always buy the food
mass-produced in a factory somewhere?LLMs can generate a million paintings that look like they are created by an expert artist. Would you visit a gallery with these paintings? Why not?
A robot can outperform any athlete. Would you rather watch the human do sports or the robot do sports? Why?
3/9
> Programs must be written for people to read, and only incidentally for machines to execute
Harold Abelson, Structure and Interpretation of Computer Programs
> Programming is the art of abstraction, of understanding how to compose smaller indeas into bigger ones, of how to understand the automation of a larger system by understanding the rules that automate smaller ones and understanding how to combine them.
https://blog.glyph.im/2026/10/programming-isnt-special.html
4/9
-
> Programs must be written for people to read, and only incidentally for machines to execute
Harold Abelson, Structure and Interpretation of Computer Programs
> Programming is the art of abstraction, of understanding how to compose smaller indeas into bigger ones, of how to understand the automation of a larger system by understanding the rules that automate smaller ones and understanding how to combine them.
https://blog.glyph.im/2026/10/programming-isnt-special.html
4/9
> Artists understand that AI is bad for art. Why don’t programmers understand that we are artists?
https://www.aitrainingstatement.org/
4.5/9
-
> Programs must be written for people to read, and only incidentally for machines to execute
Harold Abelson, Structure and Interpretation of Computer Programs
> Programming is the art of abstraction, of understanding how to compose smaller indeas into bigger ones, of how to understand the automation of a larger system by understanding the rules that automate smaller ones and understanding how to combine them.
https://blog.glyph.im/2026/10/programming-isnt-special.html
4/9
Don't take this from me. Take this from 28 people who have been awarded the Fields Medal in mathematics.
> But solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight. Forgetting this in the world of AI may turn the tool against the primary goal. Indeed, the mass production at faster and faster pace of "true/false" statements could destroy fertile ground instead of breathing life into new ideas.
5/9
-
Don't take this from me. Take this from 28 people who have been awarded the Fields Medal in mathematics.
> But solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight. Forgetting this in the world of AI may turn the tool against the primary goal. Indeed, the mass production at faster and faster pace of "true/false" statements could destroy fertile ground instead of breathing life into new ideas.
5/9
> Using 30 months of panel data on 26,811 Chinese students in grades 7-12, we study how generative AI affects homework productivity and learning. [...] AI adoption raises homework scores by 18% and reduces completion time by 30%, but lowers monthly exam scores by 20% within six months. High-stakes entrance-exam scores fall by 18 and 24%,
"The Generative AI Learning Penalty: Evidence from Chinese Secondary Education" by David Strömberg, Victor Lei, Yanhui Wu https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6868618
6/9
-
> Using 30 months of panel data on 26,811 Chinese students in grades 7-12, we study how generative AI affects homework productivity and learning. [...] AI adoption raises homework scores by 18% and reduces completion time by 30%, but lowers monthly exam scores by 20% within six months. High-stakes entrance-exam scores fall by 18 and 24%,
"The Generative AI Learning Penalty: Evidence from Chinese Secondary Education" by David Strömberg, Victor Lei, Yanhui Wu https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6868618
6/9
A graph from the paper "The Generative AI Learning Penalty: Evidence from Chinese Secondary Education" by David Strömberg, Victor Lei, Yanhui Wu. The graph shows how students who who were using genAI scored much higher in their homeworks but much lower in their exams. For students who didn't use genAI, one can see an about linear dependency between homework scores and exam scores in the graph.
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6868618
7/9
-
A graph from the paper "The Generative AI Learning Penalty: Evidence from Chinese Secondary Education" by David Strömberg, Victor Lei, Yanhui Wu. The graph shows how students who who were using genAI scored much higher in their homeworks but much lower in their exams. For students who didn't use genAI, one can see an about linear dependency between homework scores and exam scores in the graph.
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6868618
7/9
> We study the impacts of AI coding assistants and agents on software engineering work, using a novel proprietary dataset from an engineering analytics platform, covering 300 million work events -- including GitHub coding activity, Jira issues, and Google Calendar events -- across 718 firms. [...]
"Artificial Intelligence in the Firm: Bottlenecks in Software Production" by Fiona Chen and James Stratton, Harvard University https://fion.ac/jellyfish.pdf
8/9
-
> We study the impacts of AI coding assistants and agents on software engineering work, using a novel proprietary dataset from an engineering analytics platform, covering 300 million work events -- including GitHub coding activity, Jira issues, and Google Calendar events -- across 718 firms. [...]
"Artificial Intelligence in the Firm: Bottlenecks in Software Production" by Fiona Chen and James Stratton, Harvard University https://fion.ac/jellyfish.pdf
8/9
> Both technologies increase coding productivity. However, productivity gains do not fully pass through to changes in software output or employment. For AI agents, this incomplete pass-through reflects a bottleneck from code review: review times increase, a larger share of code updates require revisions, and reviews involve more comments.
8.5/9
-
> We study the impacts of AI coding assistants and agents on software engineering work, using a novel proprietary dataset from an engineering analytics platform, covering 300 million work events -- including GitHub coding activity, Jira issues, and Google Calendar events -- across 718 firms. [...]
"Artificial Intelligence in the Firm: Bottlenecks in Software Production" by Fiona Chen and James Stratton, Harvard University https://fion.ac/jellyfish.pdf
8/9
The best thing about computers is that they will always do the same thing given
the same input. They cannot do anything else because what they do is determined by an algorithm.Did software become better, more reliable or more secure now that so many started using LLMs? No, now we just have more badly maintained software.
9/9
-
The new semester started and this year I have added nine slides for my intro lecture in C++ about the use of LLMs when trying to get better at programming.
It's not on the slides but I may have mentioned the thing with fascist billionaires and pollution of our planet in my verbal commentary on the topic.
No ALT text for the attached PNG renderings of the slides. The text in the images is the text of the post itself.
This is a thread with my nine slides I presented yesterday.
@josch BIG APPLAUSE, the first slide nails it absolutely!
-
A anderslund@expressional.social shared this topic
