This was very thoughtful essay; I've had similar lines of thinking myself. How will we debug auto-generated AI code?
That said, on that topic, the essay overlooks one key point and line of reasoning about debugging, one derived from Kernighan's Law. (Kernighan as in the K in AWK + K&R C...)
The law states: "Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it."
If Kernighan's law is "true" in some rough sense (as I have long agreed with), then we have a potential solution to the "AI debugging" problem... ask the LLM to make the code four times simpler than it needs to or write code with a 4x dumber model. Then a smarter model (or us) can debug it. Right?
slightly off-topic (but tangentially mentioned in the article). I google'd "Chicxulub" and saw an easter-egg I hadn't previously known about :)
Great article. I recently finished my second reading of TMM and how/to what extent our current era of generative code affects the ideas of the book was top of mind.
What if the large tech companies systems are actually just old and we don't need that many people to maintain them and change them. that way they don't have to admit the technology companies are a mature industry and their profits about about to get competed away.
I wondered recently how many saas systems could get replaced by a database with a chat front end. is saas going to go the way of cli which was enter by a mouse and gui which was eaten by the web/saas?
if so how much do we need to maintain to keep that kind of system up an running.
A whole lotta words and not a single mention of the actual book referenced in the title. What’s the paradox? How does it relate to the Mythical Man Month? Who knows!
The Mythical Machine-Month Paradox – How much could AI change programming?
(tucson-josh.com)25 points by tucson-josh 20 hours ago | 11 comments
Comments
That said, on that topic, the essay overlooks one key point and line of reasoning about debugging, one derived from Kernighan's Law. (Kernighan as in the K in AWK + K&R C...)
The law states: "Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it."
If Kernighan's law is "true" in some rough sense (as I have long agreed with), then we have a potential solution to the "AI debugging" problem... ask the LLM to make the code four times simpler than it needs to or write code with a 4x dumber model. Then a smarter model (or us) can debug it. Right?
Great article. I recently finished my second reading of TMM and how/to what extent our current era of generative code affects the ideas of the book was top of mind.
if so how much do we need to maintain to keep that kind of system up an running.