Great post, so happy to see you writing again! When will people learn that faster doesn't equal better, and that AI slop is just not it and never will be? I literally can't even engage with AI-generated writing, and the difference against thoughtful human writing is so stark it's comical.
I think you are partly right, but also partly wrong.
First, regarding vibe coding etc. Let me tell you an anecdote of how I spent parts of several days last week. $dayjob had to show something off at a conference last weekend in San Francisco. The Sales guy had an idea and vibe coded it because the rest of us were busy. Well actually he vibe coded about 80% and then got stuck and I was volunteered to fix it. Part of the vibe coded stuff was excellent. Part of it was WTF. And the bit he'd got stuck on was not something I could get an AI to do for me either. BUT I could ask the AI for snippets and that massively shortened the amount of time I had to spend developing. This wasn't a big task - it was a fairly straight forward web page and some moderately tricky python on the back end to generate the data it needed. AI was generally speaking a massive help in that the sales guy could not have created what he wanted in look and feel without it. AI also massively helped me with python and javascript snippets that I could have written but AI got right first try in 20 seconds whereas it would take me several minutes and I'd likely have typos.
Verdit: AI massive help as an ASSISTANT. But not up to writing complete apps on its own.
Same applies, IMHO, to art, books and substack articles. I haven't read Mat Goodwin's book because I know about 90% of what it says so why bother? but I can certainly see AI being useful for bulking out verbiage and digging up citations, the catch of course with the latter is you have to read said citations and confirm that a) they exist and b) say what AI thinks they say. An example of someone who uses AI to write substack articles mostly successfully is Mark Atwood and it certainly allows him to churn out many more articles than he could if he wrote the whole thing. The articles have certain tics (load-bearing seems to be a favorite AI word I think)
Yes there are purists who refuse anything that has been touched by AI but I think there's a practically oriented set who accept AI as a valuable tool. That includes me. AI can help. But you have to watch it like a hawk - or as one of my acquaintances put it: you can trust him just as much as you can trust Sanjay your outsourced offshored consultant. Once you get the appropriate mindset then AI generated product is no less valid that product generated by Sanjay and friends, with similar tics in language and so on.
I don't reject stuff that originates via AI, but if I can tell it has AI I have less trust in it, not necessarily because the AI is likely to be wrong, but because when AI leaks through it indicates potentially excessive reliance on it by the purported creator and that suggests something slapped together quickly as opposed to something of quality
Yes. Though secretary may be the wrong word. Secretaries are well known to run orgnizations by controlling access and the like. You don't want your LLM running your life or organization
"An example of someone who uses AI to write substack articles mostly successfully is Mark Atwood and it certainly allows him to churn out many more articles than he could if he wrote the whole thing."
I suppose my own attitude toward this is the one Matthew Hughes describes in this piece: as soon as I find out someone has used genAI in their writing, I don't want to read it. Maybe this is unfair and predjudicial on my part, but I figure that if a person didn't choose to write those words, then why should I choose to read them?
I also agree with your reasoning that AI use is associated with lower quality and lower care, which is a more practical objection than mine. I go a step further and say that, even if I would have otherwise liked the article, I still don't want to read it if I know AI played any non-trivial role.
The line about catching the scent of AI is on point; the better-models point is a part folks skip. A cleaner model scrubs the quality smell. It does nothing about the deeper issue: it don't smell like home cooking 'cuz ain't nobody home.
A better model might fix the double-left armchair. It can't fix that there's no one seated in it. People flinch because some part of them can smell the absence and no upgrade's going to ship a SELF (at least in the current architecture.)
So yeah, revulsion is a good compass. Where I'm less onboard is the hope that it all dies on its own. It's a pretty good bet the current economics collapses but the web didn't disappear after the dot.com bust and there are all these open source LLMs out there for sloppers to use. So it might just mean we end up with a long period of... more of this.
The thing that doesn't ride on AI conveniently dying is the boring stuff: consent, labeling, owning your own work. The LLM-Men may try that play again.
The smell won't save us... but maybe it's telling us what to protect?
Yeah, and we’ve spent $1tn on buses and subways in the past few years, and buses and subways only have a useful life of (give or take) six years, like a GPU does.
Great post, so happy to see you writing again! When will people learn that faster doesn't equal better, and that AI slop is just not it and never will be? I literally can't even engage with AI-generated writing, and the difference against thoughtful human writing is so stark it's comical.
I think you are partly right, but also partly wrong.
First, regarding vibe coding etc. Let me tell you an anecdote of how I spent parts of several days last week. $dayjob had to show something off at a conference last weekend in San Francisco. The Sales guy had an idea and vibe coded it because the rest of us were busy. Well actually he vibe coded about 80% and then got stuck and I was volunteered to fix it. Part of the vibe coded stuff was excellent. Part of it was WTF. And the bit he'd got stuck on was not something I could get an AI to do for me either. BUT I could ask the AI for snippets and that massively shortened the amount of time I had to spend developing. This wasn't a big task - it was a fairly straight forward web page and some moderately tricky python on the back end to generate the data it needed. AI was generally speaking a massive help in that the sales guy could not have created what he wanted in look and feel without it. AI also massively helped me with python and javascript snippets that I could have written but AI got right first try in 20 seconds whereas it would take me several minutes and I'd likely have typos.
Verdit: AI massive help as an ASSISTANT. But not up to writing complete apps on its own.
Same applies, IMHO, to art, books and substack articles. I haven't read Mat Goodwin's book because I know about 90% of what it says so why bother? but I can certainly see AI being useful for bulking out verbiage and digging up citations, the catch of course with the latter is you have to read said citations and confirm that a) they exist and b) say what AI thinks they say. An example of someone who uses AI to write substack articles mostly successfully is Mark Atwood and it certainly allows him to churn out many more articles than he could if he wrote the whole thing. The articles have certain tics (load-bearing seems to be a favorite AI word I think)
Yes there are purists who refuse anything that has been touched by AI but I think there's a practically oriented set who accept AI as a valuable tool. That includes me. AI can help. But you have to watch it like a hawk - or as one of my acquaintances put it: you can trust him just as much as you can trust Sanjay your outsourced offshored consultant. Once you get the appropriate mindset then AI generated product is no less valid that product generated by Sanjay and friends, with similar tics in language and so on.
I don't reject stuff that originates via AI, but if I can tell it has AI I have less trust in it, not necessarily because the AI is likely to be wrong, but because when AI leaks through it indicates potentially excessive reliance on it by the purported creator and that suggests something slapped together quickly as opposed to something of quality
I think this mindset is what is going to win out, personally. AI is a fantastic secretary but definitely no specialist.
Yes. Though secretary may be the wrong word. Secretaries are well known to run orgnizations by controlling access and the like. You don't want your LLM running your life or organization
"An example of someone who uses AI to write substack articles mostly successfully is Mark Atwood and it certainly allows him to churn out many more articles than he could if he wrote the whole thing."
I suppose my own attitude toward this is the one Matthew Hughes describes in this piece: as soon as I find out someone has used genAI in their writing, I don't want to read it. Maybe this is unfair and predjudicial on my part, but I figure that if a person didn't choose to write those words, then why should I choose to read them?
I also agree with your reasoning that AI use is associated with lower quality and lower care, which is a more practical objection than mine. I go a step further and say that, even if I would have otherwise liked the article, I still don't want to read it if I know AI played any non-trivial role.
PS this metaphor about locomotives and bicycles is very very good - https://persuasion1.substack.com/p/what-the-inuit-know-and-ai-doesnt?r=7yrqz&utm_campaign=post-expanded-share&utm_medium=post%20viewer
Apology accepted, very happy to see you posting!
The line about catching the scent of AI is on point; the better-models point is a part folks skip. A cleaner model scrubs the quality smell. It does nothing about the deeper issue: it don't smell like home cooking 'cuz ain't nobody home.
A better model might fix the double-left armchair. It can't fix that there's no one seated in it. People flinch because some part of them can smell the absence and no upgrade's going to ship a SELF (at least in the current architecture.)
So yeah, revulsion is a good compass. Where I'm less onboard is the hope that it all dies on its own. It's a pretty good bet the current economics collapses but the web didn't disappear after the dot.com bust and there are all these open source LLMs out there for sloppers to use. So it might just mean we end up with a long period of... more of this.
The thing that doesn't ride on AI conveniently dying is the boring stuff: consent, labeling, owning your own work. The LLM-Men may try that play again.
The smell won't save us... but maybe it's telling us what to protect?
Buses and subways aren’t cool either. But they are very useful and they aren’t going away because a lot of people whine about them.
Yeah, and we’ve spent $1tn on buses and subways in the past few years, and buses and subways only have a useful life of (give or take) six years, like a GPU does.
The real tell will be what the children think of it. If they pretend to hate it because their parents do, but adapt it, game over.
If they go for it as a means of rebellion, that’d be kinda funny actually.