Imagine Knuth's heartbreak when he sees how LLMs have perverted the practical application of the art of computer programming. ("The LLM understands so I don't have to.") It's sad it happened during his lifetime. Has he commented on the topic? Anyone have a link?
> I find it fascinating that novelists galore have written for decades about scenarios that might occur after a "singularity" in which superintelligent machines exist. But as far as I know, not a single novelist has realized that such a singularity would almost surely be preceded by a world in which machines are 0.01% intelligent (say), and in which millions of real people would be able to interact with them freely at essentially no cost.
> I myself shall certainly continue to leave such research to others, and to devote my time to developing concepts that are authentic and trustworthy. And I hope you do the same.
He is still alive (I think?) you could just ask him. I doubt he is sad as much as he is excited. Computer scientists are not SWEs worried about losing their careers.
"I myself shall certainly continue to leave such research to others,
and to devote my time to developing concepts that are authentic
and trustworthy. And I hope you do the same. Best regards, Don"
There's more than one cherry to pick if one needs Mr. Knuth to have a purely-negative opinion about LLMs, but naturally any fascination is offset by the same concerns that any sane technologist has. In any case, it's all in his post.
The techno pessimists on HN are probably not PhDs in computer science. I don’t think they understand what it takes to get there, and how it shapes your thinking afterwards.
Neither Wolf nor Knuth are PhDs in Computer Science, yet many would agree that both understand "what it takes to get there" as do many others who else live sans a PhD in Comp. Sci.
Hopefully some are visionary enough to be dismayed that the endgame of their field is the acceleration of slop and fraud, the end of customer service, and the end of the reading of full, original documents.
I can't imagine being excited about any of that unless I was trying to make money from it.
While I can't speak for Knuth, I have been reflecting on the fact that developing with a modern LLM seems to be an evolution of the concept of Literate Programming that Knuth has long been a proponent of.
What is the rationale behind the assertion that Knuth would be so fundamentally opposed to the use of LLMs in development?
In literate programming you meticulously write code (as usual) but present it to a human reader as an essay: as a web of code chunks connected together in a well-defined manner with plenty of informal comments describing your thinking process and the "story" of the program. You write your program but also structure it for other humans to read and to understand.
LLM software development tends to abandon human understanding. It tends to abandon tight abstractions that manage complexity. It abandons the trappings of human software engineering.
> …LLMs have perverted the practical application of the art of computer programming. ("The LLM understands so I don't have to.") It's sad it happened during his lifetime.
If you see magazines articles or TV shows and ads from the 1980s (a fun rabbit hole on YouTube, like the BBC Archive), the general promise was that "Computers can do anything, if you just program them."
Well, nobody could figure out how to program them. (except the outcasts like us who went on to suffer for the rest of our lives for it :')
OS makers like Microsoft/Apple/etc all had their own ideas about how we should make apps and none of them wanted to work together and still don't.
With phones & "AI" everywhere we are actually closer to that original promise of everyone having a computer and being able to do anything with it, that isn't solely dictated by corporations and their prepackaged apps:
Ideally ChatGPT etc should be able to create interactive apps on the fly on your iPhone etc. Imagine having a specific need and just being able to say it and get a custom app right away just for you on your device.
What a weird, bitchy article. Knuth might be wrong but I gave up.
Imagine Knuth's heartbreak when he sees how LLMs have perverted the practical application of the art of computer programming. ("The LLM understands so I don't have to.") It's sad it happened during his lifetime. Has he commented on the topic? Anyone have a link?
https://cs.stanford.edu/~knuth/chatGPT20.txt is a conversation between Knuth and Wolfram about GPT-4.
> I find it fascinating that novelists galore have written for decades about scenarios that might occur after a "singularity" in which superintelligent machines exist. But as far as I know, not a single novelist has realized that such a singularity would almost surely be preceded by a world in which machines are 0.01% intelligent (say), and in which millions of real people would be able to interact with them freely at essentially no cost.
> I myself shall certainly continue to leave such research to others, and to devote my time to developing concepts that are authentic and trustworthy. And I hope you do the same.
That's related, thank you for posting it. But what does Knuth think of "vibe coding" or "agentic coding"? Apparently he hasn't commented yet.
I don't know why but it makes me smile that he did this experiment by having a grad student type the questions for chatgpt and copy the results.
He is still alive (I think?) you could just ask him. I doubt he is sad as much as he is excited. Computer scientists are not SWEs worried about losing their careers.
He’s still here. In fact, in December he gave his annual Christmas lecture, and last month he was a guest at a Computer History Museum event.
Excited? I doubt that. I'm guessing you haven't read his books.
He seems pretty fascinated with the possibilities.
https://cs.stanford.edu/~knuth/chatGPT20.txt
"I myself shall certainly continue to leave such research to others, and to devote my time to developing concepts that are authentic and trustworthy. And I hope you do the same. Best regards, Don"
There's more than one cherry to pick if one needs Mr. Knuth to have a purely-negative opinion about LLMs, but naturally any fascination is offset by the same concerns that any sane technologist has. In any case, it's all in his post.
The techno pessimists on HN are probably not PhDs in computer science. I don’t think they understand what it takes to get there, and how it shapes your thinking afterwards.
Neither Wolf nor Knuth are PhDs in Computer Science, yet many would agree that both understand "what it takes to get there" as do many others who else live sans a PhD in Comp. Sci.
Hopefully some are visionary enough to be dismayed that the endgame of their field is the acceleration of slop and fraud, the end of customer service, and the end of the reading of full, original documents.
I can't imagine being excited about any of that unless I was trying to make money from it.
Amen.
While I can't speak for Knuth, I have been reflecting on the fact that developing with a modern LLM seems to be an evolution of the concept of Literate Programming that Knuth has long been a proponent of.
What is the rationale behind the assertion that Knuth would be so fundamentally opposed to the use of LLMs in development?
I don't see the connection.
In literate programming you meticulously write code (as usual) but present it to a human reader as an essay: as a web of code chunks connected together in a well-defined manner with plenty of informal comments describing your thinking process and the "story" of the program. You write your program but also structure it for other humans to read and to understand.
LLM software development tends to abandon human understanding. It tends to abandon tight abstractions that manage complexity. It abandons the trappings of human software engineering.
> …LLMs have perverted the practical application of the art of computer programming. ("The LLM understands so I don't have to.") It's sad it happened during his lifetime.
If you see magazines articles or TV shows and ads from the 1980s (a fun rabbit hole on YouTube, like the BBC Archive), the general promise was that "Computers can do anything, if you just program them."
Well, nobody could figure out how to program them. (except the outcasts like us who went on to suffer for the rest of our lives for it :')
OS makers like Microsoft/Apple/etc all had their own ideas about how we should make apps and none of them wanted to work together and still don't.
With phones & "AI" everywhere we are actually closer to that original promise of everyone having a computer and being able to do anything with it, that isn't solely dictated by corporations and their prepackaged apps:
Ideally ChatGPT etc should be able to create interactive apps on the fly on your iPhone etc. Imagine having a specific need and just being able to say it and get a custom app right away just for you on your device.
Past progress in software engineering is a tower of well-defined abstractions.
Compilers for languages that make specific guarantees about the semantics of their translation to machine code.
Libraries with well-defined interfaces that let you stand on the shoulders of others by understanding said interfaces and ignoring the internals.
This is how concrete progress is made. You build on solid blocks.
That era is ending.
Title should note that this is a 2015 post.