> LLMs, by definition, can’t make anything thoughtfully, because LLMs have no thoughts. ... apps that feel crafted, that show refinement, that feel intuitive—requires a human developer not only with taste, but with understanding.
Well said, thank you.
It saddens me that people can't understand this. Already there is the fallacy of "I can make apps with AI, therefore no one needs to code anymore," as if all coding was was making apps. But even the apps that people think are passable feel so fake, clunky and sludgy.
LLMs will give responses based on data that already exists. There's plenty of data that allows an LLM to write working code - possibly even high-quality code! But there's not much data that says how to make a "great" app, I think because it's domain specific.
Consider:
What made Firefox great was the lived experience of people fed up with IE6. The Internet talked plenty about speed, tabs, and stability, but "speed, tabs, and stability" undersells much of what made Firefox stand out.
What made Ocarina of Time great was that it defined a genre. No one really knew yet whether the story, the graphics, the dungeon design, or the marketing was going to sell that game. You couldn't yet ask any other Miyamotos what had worked for a 3D adventure.
We still arguing over what made Tom Brady (the American football player) great, not being able to nail down if it was the length of his career, the number of times he won the Super Bowl, his individual game records, his coach, the rest of his team, or something else entirely.
I think, likewise, the criteria for a "great" app aren't known until the app is already great, and that gives an LLM no specific advantage.
The "problem" (if it is one) is the same as we had with previous craftsmen being replaced by mass-production. Since it doesn't "have" to be done by hand, it can be done quickly and cookie-cutter.
And the dirty secret is that 90% of "stuff" is crud already (literally CRUD or just a tool for a task, no thought needed).
Everyone loves those little apps that used to be so common on the Mac, lovingly created and amazingly excellent at some small task; but the vast majority of people use whatever tool barely works and move on with their life. AI is incredibly enabling there; even if all you ever use it for is automating away spreadsheets.
Nah. In the big picture, voice can only ever be a complementary interface, not a widespread substitute.
It's fundamentally some combination of slow, imprecise, and ambiguous for most practical tasks and more importantly it places heavy demands on both the environment and the capabilities of the user.
It can be profoundly satisfying or efficient when the stars align over those limitations, but in the scheme of what we want computer-operated tools to do, these scenarios ends up covering a very narrow set of use cases.
The ultimate problem with voice interfaces wasn't technological, but biological/sociological: most people are simply inarticulate and impatient. They don't have the capacity to express themselves well in speech and don't have the patience to trudge through misunderstanding and clarification cycles -- especially when they lack the reflective insight to understand why they're being misunderstood. That's why they're not public speakers or writers or counselors or programmers or such (and even many of them are poorly suited). So even while the most limiting technological barriers may have been cracked open now, we'll still have to produce tool controls that accommodate these inarticulate, impatient people.
A part of software UX design is instrumental interaction. I could replace my guitar playing with singing (or asking a model to play the right notes). But that won't give me the feeling of fingers against the strings, or of moving effortlessly across the fretboard. There's a fair amount of UX use that follows the same idea, where the software becomes an instrument of sorts: https://link.springer.com/rwe/10.1007/978-3-319-27648-9_23-1
I think different parts of systems carry different levels of risk. Some time ago I proposed a model for assessing the risk in a granular way. Putting a value (CCD - Cost of Cognitive Debt) on individual parts may foster healthy discussions within teams. Elaborating on it here:
For awards, the use of A.I. should be baked into the category descriptions/rules of eligibility up-front, and the organizers should be free to have a contest which is AI-free (in terms of contestants, I'd like to hope bots would always be banned from voting), or a contest which allows it, or separate contest categories for use and non-use.
For my part, I'm late to the game --- I used Google's LLM-enabled notebook to generate some sample DXF files early on, but I mostly hack away and copy-paste from Stackexchange and so forth --- flet.dev hitting 1.0 and offering an online studio did have me buying tokens and ultimately a subscription to make:
while waiting for my next month of tokens to show up by copy-pasting and in a couple of instances asking Copilot about errors.
Currently trying to work out how to get the previews in the first app to look right, maybe adding an optional toolpath preview to the latter, and how to get a larger and more complex version up-and-running with an architecture I can wrap my head around using a toolchain I feel comfortable with (rather than a browser app, my typical development chain is local driven by a .text file: https://github.com/WillAdams/gcodepreview using a Literate Programming system).
The weird line is for knowledge lookup, which is sadly becoming more and more the case to make up for the very intentional enshittification of 10's era search tools. Given the recent walking elephant(-ish thingy) in the room for games, would I be using Ai if I consulted to claude on an issue and saw (didn't directly use, but saw) a snippet of code? Many would say no, but the environmental angle would suggest I am still contributing to AI pollution and the ethical side would suggest I am ultimately relying on a fountain of unlicensed code to get that answer.
But if you think that; how would you begin to enforce such matters? There wouldn't be any trace of AI contributions in a repo. Some theoretical perfect AI detector wouldn't detect anything because I synthesized the scripted into my own mind and figured out an approach to fix my problem. But the "damage" was done, nonetheless. That's what this current situation is having me consider.
For wear reasons alone, I don't see myself typing stuff in full with the one set of hands I have, when I can also make the machine do it.
RSI (Repetitive strain injury) is no joke in our profession.
So that alone makes the moral 100% anti AI standpoint a bit weak and surprisingly ignorant, given that a lot of the anger is (rightfully) about ignorant tech bros bulldozing reality with slop.
What sort of ego leads a person who doesn't use these tools to write articles about their capabilities like this?
> Anecdotal tales of discovering Claude has written four versions of the same function in different places, finding giant sections of code that never get called, and “bug-finding” agents that imagine and even create bugs are growing rampant.
Growing rampant? As in their getting worse? Because that's not at all my experience.
> The LLMs themselves don’t even “know” they’re working with code.
lol What does this even mean? What does "know" mean in this context? Models identify the code, the language its written in, the domain you're targeting, etc.
Reading stuff like this, I mentally recoil from the tone of self-righteous moral outrage. The ethical issues around AI don't seem any different than the rest of modern capitalism. But things like industrialised farming, manufacturing, fossil fuels, chemical production, etc. don't seem to generate anything close to the quasi-religious fervour that hating AI inspires in certain people.
It's difficult to escape the conclusion that the difference is that the harm from those other industries mostly accrues to the poor and the working class, whereas AI is threatening the livelihoods of rich, educated westerners who thought their social status insulated them from economic shocks like this.
No, they do generate that outrage, you've just become used to ignoring it.
Besides, outrage is mostly a response to change, or seeing something in the news (a major outrage driver). If there's no news item to hang it on the outrage just sort of sits there uselessly.
There's a lot of outrage to go round. An entire US administration devoted to maximising gross national outrage will do that.
The class commentary affects why you're hearing about it, not what the baseline level of outrage is.
> The ethical issues around AI don't seem any different than the rest of modern capitalism.
AI makes copyright and patents obsolete. There are no protections left for original nor derivative works; AI will hoover it all up and reproduce it without qualm and at scale. It must be trained, and it can only be trained on what already exists.
> But things like industrialised farming, manufacturing, fossil fuels, chemical production, etc. don't seem to generate anything close to the quasi-religious fervour that hating AI inspires in certain people.
A computer is a general-purpose machine. A tractor is not. AI is threatening every job that involves using a computer to accomplish it, or that could use computers and automation to accomplish it. Not to mention the promise of exponential growth; AI will continue to improve and in turn be used to improve itself, which a tractor cannot do.
These kinds of threats are what inspire such fears.
> The ethical issues around AI don't seem any different than the rest of modern capitalism.
Let's fix that while we're at it too. I'm a bit tired of the "we already ruined this much. What's one more broken window" handwave. Let's start fixing windows instead.
>It's difficult to escape the conclusion that the difference is that the harm from those other industries mostly accrues to the poor and the working class, whereas AI is threatening the livelihoods of rich, educated westerners who thought their social status insulated them from economic shocks like this.
If you somehow assume that these people never cared about fossil fuels, manufacturing, unethical farming, etc. Yes. It can be easy to make that conclusion. Yet somehow, those have all been hot button issues for 30+ years at this point. Maybe they simply see an old pattern emerging to repeat yet again?
> LLMs, by definition, can’t make anything thoughtfully, because LLMs have no thoughts. ... apps that feel crafted, that show refinement, that feel intuitive—requires a human developer not only with taste, but with understanding.
Well said, thank you.
It saddens me that people can't understand this. Already there is the fallacy of "I can make apps with AI, therefore no one needs to code anymore," as if all coding was was making apps. But even the apps that people think are passable feel so fake, clunky and sludgy.
LLMs will give responses based on data that already exists. There's plenty of data that allows an LLM to write working code - possibly even high-quality code! But there's not much data that says how to make a "great" app, I think because it's domain specific.
Consider:
What made Firefox great was the lived experience of people fed up with IE6. The Internet talked plenty about speed, tabs, and stability, but "speed, tabs, and stability" undersells much of what made Firefox stand out.
What made Ocarina of Time great was that it defined a genre. No one really knew yet whether the story, the graphics, the dungeon design, or the marketing was going to sell that game. You couldn't yet ask any other Miyamotos what had worked for a 3D adventure.
We still arguing over what made Tom Brady (the American football player) great, not being able to nail down if it was the length of his career, the number of times he won the Super Bowl, his individual game records, his coach, the rest of his team, or something else entirely.
I think, likewise, the criteria for a "great" app aren't known until the app is already great, and that gives an LLM no specific advantage.
The "problem" (if it is one) is the same as we had with previous craftsmen being replaced by mass-production. Since it doesn't "have" to be done by hand, it can be done quickly and cookie-cutter.
And the dirty secret is that 90% of "stuff" is crud already (literally CRUD or just a tool for a task, no thought needed).
Everyone loves those little apps that used to be so common on the Mac, lovingly created and amazingly excellent at some small task; but the vast majority of people use whatever tool barely works and move on with their life. AI is incredibly enabling there; even if all you ever use it for is automating away spreadsheets.
It seems like almost every argument defending the slide into agentic development is either:
1. This is just like the move to [X deterministic technology]. (False analogy)
2. You think it was good before? Who cares if it's a bit worse. (Corruption mentality doomerism)
Funny thing is that people still use AI to write GUI apps, while in the future the entire app can often be replaced by a voice interface.
Nah. In the big picture, voice can only ever be a complementary interface, not a widespread substitute.
It's fundamentally some combination of slow, imprecise, and ambiguous for most practical tasks and more importantly it places heavy demands on both the environment and the capabilities of the user.
It can be profoundly satisfying or efficient when the stars align over those limitations, but in the scheme of what we want computer-operated tools to do, these scenarios ends up covering a very narrow set of use cases.
The ultimate problem with voice interfaces wasn't technological, but biological/sociological: most people are simply inarticulate and impatient. They don't have the capacity to express themselves well in speech and don't have the patience to trudge through misunderstanding and clarification cycles -- especially when they lack the reflective insight to understand why they're being misunderstood. That's why they're not public speakers or writers or counselors or programmers or such (and even many of them are poorly suited). So even while the most limiting technological barriers may have been cracked open now, we'll still have to produce tool controls that accommodate these inarticulate, impatient people.
A part of software UX design is instrumental interaction. I could replace my guitar playing with singing (or asking a model to play the right notes). But that won't give me the feeling of fingers against the strings, or of moving effortlessly across the fretboard. There's a fair amount of UX use that follows the same idea, where the software becomes an instrument of sorts: https://link.springer.com/rwe/10.1007/978-3-319-27648-9_23-1
That sounds like hell.
I think different parts of systems carry different levels of risk. Some time ago I proposed a model for assessing the risk in a granular way. Putting a value (CCD - Cost of Cognitive Debt) on individual parts may foster healthy discussions within teams. Elaborating on it here:
https://epiq.sh/blog/cost-of-cognitive-debt
For awards, the use of A.I. should be baked into the category descriptions/rules of eligibility up-front, and the organizers should be free to have a contest which is AI-free (in terms of contestants, I'd like to hope bots would always be banned from voting), or a contest which allows it, or separate contest categories for use and non-use.
For my part, I'm late to the game --- I used Google's LLM-enabled notebook to generate some sample DXF files early on, but I mostly hack away and copy-paste from Stackexchange and so forth --- flet.dev hitting 1.0 and offering an online studio did have me buying tokens and ultimately a subscription to make:
https://studio.flet.dev/apps/bicQWc3Aul
but for a long while it wasn't running and I was out of tokens, so I coded up:
https://studio.flet.dev/apps/46tp8GPq9g
while waiting for my next month of tokens to show up by copy-pasting and in a couple of instances asking Copilot about errors.
Currently trying to work out how to get the previews in the first app to look right, maybe adding an optional toolpath preview to the latter, and how to get a larger and more complex version up-and-running with an architecture I can wrap my head around using a toolchain I feel comfortable with (rather than a browser app, my typical development chain is local driven by a .text file: https://github.com/WillAdams/gcodepreview using a Literate Programming system).
The weird line is for knowledge lookup, which is sadly becoming more and more the case to make up for the very intentional enshittification of 10's era search tools. Given the recent walking elephant(-ish thingy) in the room for games, would I be using Ai if I consulted to claude on an issue and saw (didn't directly use, but saw) a snippet of code? Many would say no, but the environmental angle would suggest I am still contributing to AI pollution and the ethical side would suggest I am ultimately relying on a fountain of unlicensed code to get that answer.
But if you think that; how would you begin to enforce such matters? There wouldn't be any trace of AI contributions in a repo. Some theoretical perfect AI detector wouldn't detect anything because I synthesized the scripted into my own mind and figured out an approach to fix my problem. But the "damage" was done, nonetheless. That's what this current situation is having me consider.
Yes, 0.
For wear reasons alone, I don't see myself typing stuff in full with the one set of hands I have, when I can also make the machine do it.
RSI (Repetitive strain injury) is no joke in our profession.
So that alone makes the moral 100% anti AI standpoint a bit weak and surprisingly ignorant, given that a lot of the anger is (rightfully) about ignorant tech bros bulldozing reality with slop.
The amount that won't make me lose my job so I can keep providing for my family.
Can we just fast-forward to the part where the moralistic hand wringing is over, and we all accept that this is a part of software development now?
Were accountants this internally conflicted about the energy use of computers when they switched to spreadsheets?
Human slop.
What sort of ego leads a person who doesn't use these tools to write articles about their capabilities like this?
> Anecdotal tales of discovering Claude has written four versions of the same function in different places, finding giant sections of code that never get called, and “bug-finding” agents that imagine and even create bugs are growing rampant.
Growing rampant? As in their getting worse? Because that's not at all my experience.
> The LLMs themselves don’t even “know” they’re working with code.
lol What does this even mean? What does "know" mean in this context? Models identify the code, the language its written in, the domain you're targeting, etc.
Reading stuff like this, I mentally recoil from the tone of self-righteous moral outrage. The ethical issues around AI don't seem any different than the rest of modern capitalism. But things like industrialised farming, manufacturing, fossil fuels, chemical production, etc. don't seem to generate anything close to the quasi-religious fervour that hating AI inspires in certain people.
It's difficult to escape the conclusion that the difference is that the harm from those other industries mostly accrues to the poor and the working class, whereas AI is threatening the livelihoods of rich, educated westerners who thought their social status insulated them from economic shocks like this.
No, they do generate that outrage, you've just become used to ignoring it.
Besides, outrage is mostly a response to change, or seeing something in the news (a major outrage driver). If there's no news item to hang it on the outrage just sort of sits there uselessly.
There's a lot of outrage to go round. An entire US administration devoted to maximising gross national outrage will do that.
The class commentary affects why you're hearing about it, not what the baseline level of outrage is.
> The ethical issues around AI don't seem any different than the rest of modern capitalism.
AI makes copyright and patents obsolete. There are no protections left for original nor derivative works; AI will hoover it all up and reproduce it without qualm and at scale. It must be trained, and it can only be trained on what already exists.
> But things like industrialised farming, manufacturing, fossil fuels, chemical production, etc. don't seem to generate anything close to the quasi-religious fervour that hating AI inspires in certain people.
A computer is a general-purpose machine. A tractor is not. AI is threatening every job that involves using a computer to accomplish it, or that could use computers and automation to accomplish it. Not to mention the promise of exponential growth; AI will continue to improve and in turn be used to improve itself, which a tractor cannot do.
These kinds of threats are what inspire such fears.
> The ethical issues around AI don't seem any different than the rest of modern capitalism.
Let's fix that while we're at it too. I'm a bit tired of the "we already ruined this much. What's one more broken window" handwave. Let's start fixing windows instead.
>It's difficult to escape the conclusion that the difference is that the harm from those other industries mostly accrues to the poor and the working class, whereas AI is threatening the livelihoods of rich, educated westerners who thought their social status insulated them from economic shocks like this.
If you somehow assume that these people never cared about fossil fuels, manufacturing, unethical farming, etc. Yes. It can be easy to make that conclusion. Yet somehow, those have all been hot button issues for 30+ years at this point. Maybe they simply see an old pattern emerging to repeat yet again?