Doesn't that mean guaranteed failure in the market if you use taste as your guiding star? Therefore, there's really no way to enter the market anymore.
I don't know how you then ship software or monetise it...? The market will just clone it.
So... Don't share stuff publicly? Only share with closed groups? Where does this leave small-time software development...?
Maybe software moves to this after it's commoditized? Users might prefer an artisanal one vs ordering it off the shelf or using an agent themselves.
This article is unfortunately full of vague slop like this. It seems the author doesn’t trust their own taste.
I also completely disagree with the premise though - so many seem desperate to discuss the implications of a singularity which shows no sign of arriving.
LLMs have ingested the entire internet and are able to generate plausible and mostly accurate text on demand.
What is missing from thee output of LLMs: intelligence, agency, motivation, morality, discrimination and yes, taste.
And honestly: I've been tired of an age of engineers that are beat down, that only want simple, that finding the most blunt approach is the only way. It has felt tasteless. Anti-ambitious. It often is still very sensible and practical and what you should shoot for! But there are also people wandering around now trying much harder! And I am excited for those frontiers! I think it will challenge and shake the foundations that we've accepted as true, as There Is No Alternative, in exciting bold and fun ways.
I am however pretty unconvinced by the article. There's a lot that doesn't quite work for me, that isn't building the case I'd like, whose takes are off from my read.
> The output is good enough—that is the problem. Good enough is a solvent. It dissolves the reason to do better. For as long as making things was expensive, the expense did quiet work on our behalf. It rationed output.
Agreed about the rationing of output. I do think that the article though continues to show a magical thinking. That we have these things now and they autonomously do the thing. That the LLM's have solved it all.
The article itself goes on at length about how nebulous and abstract taste is:
> Taste is that. It is the compressed, wordless verdict you reach faster than you can justify. It is the “no, again” you say to yourself with total conviction and no available argument
But this contrasts so distinctly to me against "the output is good enough". Is it? That depends. That depends on your taste. The proximate first results come quick. But the technics underneath? Those themselves, in my view, rely enormously on engineering taste to support and advance. I think we see a very similar sort of magical delivery thinking, very clearly on display here:
> When the factories came, they could suddenly make everything—cheaply, uniformly, by the thousand.
As if there was some magical "good enough" transition where suddenly the aliens came and gave us this box that just does the thing. As if we discovered the right formulas and math and now: we had production. Again I think there's just an enormous amount of work and taste that is still actively required inside the factory box, that building the industrial processes is incredibly intense & difficult, even though we have reliable industry-line production and now robots doing the labor.
There's still so many gems, so much lovely material for thought. I love the provocations here, and I think there's a lot of great calibration.
> The friction was not an obstacle to developing taste. The friction was the curriculum.
Makes me think of yesterdays @apenwarr banger,
> Every slow prototype started out as a fast prototype, I think that’s how it goes https://bsky.app/profile/apenwarr.ca/post/3msemlo4rds2h
I would like to comment on this, though:
> that only want simple, that finding the most blunt approach is the only way. It has felt tasteless. Anti-ambitious.
I am always aiming for simple. If it's simple, it's maintainable and can be easy to reason about. Doing that is hard work, though. I used to spend three iterations to achieve it: functions are simple, easy to reason about, properly named and composable.
It's the opposite of blunt, though, so it might be that you meant something else with the term "simple" :)
> But the ratio was never the danger. It held steady for centuries. What held the flood back was that producing the crap cost something. Bad novels still took a year to write. Bad software still took a month to build.
I want to rename the title to "Blogs writing in LLMish are all that's left".
Seriously, take this one: "But the ratio was never the danger.".
Who the FUCK speaks like that?
Every day is a new day. It is better to be lucky. But I would rather be exact. Then when luck comes you are ready.
The reason is simple: Short sentences are punchy. They hit hard, carry emotional weight.
P.S. As evident, clearly I'm having some fun with this, but the point stands.
> The reason is simple. Short sentences are punchy. They hit hard. But they carry emotional weight.
Does this stem from you believing that there is value in making logical errors, or is your angle more towards believing that nobody should waste their time interacting with someone who makes logical errors?
Do you mean starting sentences with "But"? I do, but then again I'm not a native speaker and my writing may be nonstandard. Or do you mean short and direct sentences? In which case, Hemingway.
I'm not saying the article isn't AI slop, but I think it's not obvious from the specific sample you chose, since plenty of people online write like that, before AI.
If we assume that it is LLM generated, as suggested, apparently a lot of people, as the LLMs were trained to mimic how people speak (write).
You might find it to be unusual because you probably have never read anything written by most people. Only peculiar characters tend to get noticed.
It's a foundational model (fresh from autoregressive pretraining) that approximates the probability distribution of human texts. And, no, it's not the statistical average of how people speak. It approximates how a person who could have written a text in its context would have written the next words.
Fine-tuning, RLHF, reinforcement learning change this probability distribution. I guess, it's mostly RLHF that shapes the way LLMs write. The similarity of style is due to common providers of RLHF data.
This is likely because the individual measures being averaged like "femur length" were not independent from one-another, even where they had the benefit of being normally distributed.
In fact, I find no meaningful difference between "But no _individual_ speaks like this." and "But the ratio was never the danger.", aside from them being about different topics. They are syntactically very similar.
Walter Hartwell White, I suppose.
It's a blog post. It was written. The written word is different from the spoken.
People don't speak like: The man in black fled across the desert, and the gunslinger followed.
People don't speak like: In a hole in the ground there lived a hobbit.
In fact, I can throw this question at you:
"TFA's title is"
Who the FUCK speaks like that? Who the FUCK uses TFA's IRL OMG WTF?
Silly, isn't it?
The person you're replying to is saying that it wasn't written. If you are the author, I will believe you that it was. If you are not the author, I have no idea why you think you know more than anyone else does.