The intent paradox of AI generated code
feruchemy.bearblog.dev
The intent paradox of AI generated code
1–10 of 18 posts
Re: The intent paradox of AI generated code
#2Wrt to premature optimization, in future, AI will write the most optimized for every general and obscure cases out of the box eking out every last blood of performance from any kind of hardware.
This will usher in the era of "hypermature optimization".
Re: The intent paradox of AI generated code
#3This is an interesting perspective and certainly makes sense that the “why” always needs the human. What I think is clear for the moment is that AI coding helpers can’t deliver anything novel by themselves, although through luck they might stumble across new invention by combination of existing human thought. The question for me is, once the majority of AI code has been re-spidered and vectorized, and reused and reused, is there anything novel left?
Re: The intent paradox of AI generated code
#4Wrt to premature optimization, in future, AI will write the most optimized for every general and obscure cases out of the box eking out every last blood of performance from any kind of hardware. This will usher in the era of "hypermature optimization".
What makes you say that besides blind faith in the exponential advancement of language models?
I’ve need no evidence that AI writes particularly performant code (especially wrt to specific hardware.)
Nor have I seen any big player showing off models which were tuned for performant code generation.
Code performance work is almost always very specific to the codebase at hand and not at all general.
Re: The intent paradox of AI generated code
#5Wrt to premature optimization, in future, AI will write the most optimized for every general and obscure cases out of the box eking out every last blood of performance from any kind of hardware. This will usher in the era of "hypermature optimization".
What makes you say that besides blind faith in the exponential advancement of language models? I’ve need no evidence that AI writes particularly performant code (especially wrt to specific hardware.) Nor have I seen any big player showing off models which were tuned for performant code generation. Code performance work is almost always very specific to the codebase at hand and not at all general.
Is there a physical law which will limit AI progress? I don't have a timeline in mind but I think that is the future.
Re: The intent paradox of AI generated code
#6Earlier quoted context omitted.
What makes you say that besides blind faith in the exponential advancement of language models? I’ve need no evidence that AI writes particularly performant code (especially wrt to specific hardware.) Nor have I seen any big player showing off models which were tuned for performant code generation. Code performance work is almost always very specific to the codebase at hand and not at all general.
Is there a physical law which will limit AI progress? I don't have a timeline in mind but I think that is the future.
> Is there a physical law which will limit AI progress?
Probably the same law that limits compute power wrt to space and energy.
Look at how much energy goes into training GPT 4.5 vs it’s improvements (on our own potentially bunk benchmarks, granted)
Also like… literally nothing in the universe advances exponentially forever.
Whether or not any of that is valid, I’d like an answer to my question.
Is the only reason you think that a blind belief that AI will advance exponentially in all domains?
Re: The intent paradox of AI generated code
#7Wrt to premature optimization, in future, AI will write the most optimized for every general and obscure cases out of the box eking out every last blood of performance from any kind of hardware. This will usher in the era of "hypermature optimization".
Already seeing this happen in salesforce implementation at work.
It's pretty crazy how many problems we have are people trying to solve already solved problems
Re: The intent paradox of AI generated code
#8Earlier quoted context omitted.
Is there a physical law which will limit AI progress? I don't have a timeline in mind but I think that is the future.
> Is there a physical law which will limit AI progress? Probably the same law that limits compute power wrt to space and energy. Look at how much energy goes into training GPT 4.5 vs it’s improvements (on our own potentially bunk benchmarks, granted) Also like… literally nothing in the universe advances exponentially forever. Whether or not any of that is valid, I’d like an answer to my question. Is the only reason y…
If AI is the general purpose pattern recognition algorithm and real world has patterns then as long as data i.e. real world has juice, AI will do better. Recursive self improvement is included in the domain. The only constraint will be how long it takes to gather feedback from the real world when AI does experiments.
Re: The intent paradox of AI generated code
#9What if instead of static code our programs of the future are written/rewritten on the fly?
Imagine a simple todo application. Instead of a conditional filter on completed items the code itself changes to always and only return completed items upon request, spoken or otherwise.
Re: The intent paradox of AI generated code
#10Future code bases will contain more detailed specifications, design documents, and company guidelines and the code will become more of a byproduct, generated as needed. I already do this more and more with my LLM config files. This has already happened awhile ago with compilers replacing machine code and more recently with open source libraries replacing proprietary code. So, yes, more "intent" based documents, which represent a higher level abstraction of the problem.