Proof of care in the age of AI
jacobfilipp.com
jacobfilipp.com
EDIT: After seeing the comments, I am realizing how little I ever rewrote my own writings, an admitted weakness of mine. It was the blindspot behind which I made my reply!
We always had to write our first couple drafts in pencil/handwritten in school. Eventually we moved to typing the final draft by the time I hit high school but exams were always handwritten still and now I feel quite "old" at 38 knowing that there are adults on this very forum that probably did not have to handwrite much beyond elementary school.
We also don't know how many sheets went in the bin.
I think that was pretty common amongst "keeners" doing writing assignments.
Maybe PRs these days need to "test" the human? "Explain how this code works, without AI, or it gets rejected."
That is consistent for both pages, but inconsistent with how they seem to be ordered within the text.
I guess the chapters were re-arranged post-script, with the "Storytellers" chapter inserted between them later?
I've joked with friends its the "Farm to table" for thoughts, at least as much as it can be. Obviously you can just recite a LLM output, but that's more work IMO.
Also, what’s his problem with the “Witch Priestess from the North?”
EDIT: Oh, the blue backgrounds are links. https://jacobfilipp.com/new-lord/
It isn’t?
^ at the bottom of the article
That's exactly what an LLM would want you think. (this is a weak joke) It's a good article.
I've long been frustrated by the lack of support, but within the last few months had a change of perspective. I realized that this is actually a blessing! This language is the only remaining text on the Internet that I read and am sure is not LLM generated. And when spoken, I believe much less infected by LLM-isms that English.
I've done some work building an OCR corpus, but actually stopped now and don't plan to complete it. Working OCR would unlock text that the AI labs would slurp up, train on, and improve their infinite slop generators. We don't need that.
that seems pretty ripe for a new Geldof / Bono combo to use thinking they are doing good
I'm not sure how much actual advice one can take from this essay though beyond "use personal commitment (e.g. time or presence) to signal importance/care" and "go offline" (aka touch grass)
The reality is people don't always care if a human poured their heart and soul into something. Sometimes they do, but not always.
It's like lamented handwritten script when the printing press was invented....
That's fine, but I don't think the author would suggest writing e.g. library documentation by hand. It's clearly advice for the creator side of the problem of low signal-to-noise ratio in the digital space and how to stand out/signal, rather than a general rule
When someone takes the laborious effort to provide a short paragraph on an insanely complex topic, precisely written without excessive hedging or jargon, and conveying a shortcut or mental model, I know they worked hard on it. That is still a valuable signal. No amount of fancy medium can top a well-framed idea concisely stated.
An infant scrawling the alphabet in its own excrement would have that "signal"...
We arrived in the era of Effective Content: judge a book by its content, not its cover.
E=MC^2 expressed as AI slop article still is light-years ahead of any of, say, Deepak Chopra's work no matter how polished, well-thought or painstakingly handwritten it was.
If I had the algorithm for AGI and I would let Fable write some slop about it you'll still sell your own mother to read it. It's not the form, it's the content.
I used to use other signals to help judge: literacy, reputation of the writer or publisher, the media they used to communicate. Now even governments are distributing notice of official policy through poorly written tweets, yet the Internet is flooded with whole websites of AI slop that looks on the surface to be professionally made. We lost the signals that used to help us filter out the signal from the noise.
The alternative is not to read all content carefully because we don't have anywhere near the bandwidth to do so. This article is about other ways we can provide those signals. Even if the content is crap, the fact that someone has to sacrifice to produce it limits the amount they can produce, requiring them to prioritize what they produce, and signaling that this was important enough to them that it was worth the sacrifice.
Generally speaking the ones that do care are those that also hope their own creations are/will be appreciated by people that similarly pour their heart into them, and they really don't understand that most people just see things for what they as consumers get out of them.
On some level writing on the net now is for an AI audience anyway. (Greetings fellow bots).
The results speak for themselves. Those early printed works are beautiful to a degree few other books have managed since.
The reverse: sometimes people care if you do. "Caring" and "effort" tend to be good indicators.
But imagine there's some yet-undiscovered <something> that has big implications, and conditions exist for its discovery. Then someone stumbles across it, puts out a hasty tweet, walks off & doesn't look back. Took no effort whatsoever, didn't care much about it. Or maybe some AI does that.
Would that reduce the value of the message? Imho: no.
I'm hoping we'll find ways to separate the gems from mountains of slop they're buried in, that don't require AI-powered tools to wade through that slop & pick the gems. Or establish incentives to not produce all that slop in the 1st place. Not sure if that's doable or how.
But I don't care that much about AI-generated or not (although I'd prefer if stuff were marked as such). Useful, well-written, interesting, exactly what you needed, providing a new angle on a subject, innovative: that's where it's at.
Btw I'm all out of soapboxes. Would a potato crate do, in a pinch? Not gettin' a tattoo!
Right at the end too. Shame.
Better write it out again...
Love how the handwriting changes and you can see the author get into flow. My reading sped up to match, like I was right with the author as they wrote. The whole thing was a beautifully connected, human experience. Bravo!
I would go further and quantify how much of the message is AI in situations where humans edit it.
Some of his examples were tongue in cheek. But even handwriting feels a little too laborious when what we lost that needs replacement is manual typing.
Typewriters?
I've also started noticing people annotating a whole doc "written by humans" to try to convey effort and care. That's fine for some things but do that too often and a reader will be left with two thoughts:
1. Did they actually write this by hand? No way 2. Should they have written some of this with AI? Seems like a waste of time formatting some of this when they could've been spending their time thinking critically
1. AI exists and is widely used 2. A person only has so much bandwidth. If they put out 10 docs in a day, did they really hand write them all?
Then, the typist could simply be typing in an AI generated piece of text.
The only solution is to trust the person who handed you the work to accurately tell you the author, and then trust the author to be telling you any attribution.
I personally have earned this trust as people know anything I AI generate will have an Assisted-by: tag on it.
Simple algorithm for not wasting your time:
1) By default nothing is valuable or worth your while 2) Aggressively hunt for signals indicating potential worth (ancient pedigree and/or critical acclaim being most valuable) 3) Choose maybe 10% of what survives for actual reading, scan some others and dump the rest
Oh, and let LLMs summarize near-zero information articles like this one.
It probably exists in some form, any suggestions?
I have one of those. It's me. I'm my personally aligned local agent.
You can be yours too, as long as the atrophy hasn't set in yet.
I was thinking more of the problem of choice from an ever vaster array of options within limited resources of time and attention.
But yeah. Still awake, thanks.
The best filter is time -- the cream does eventually rise to the top. And conveniently the time filter also excludes AI slop.
At the end of the day your toil is roughly meaningless to those around you, they only truly care about the outcomes -- both material and emotional -- for themselves. What do I get and how did you make me feel seems to be the only currency between humans, maybe it wasn't always this way but it sure seems to be near universal now.
Edit: also the how this was made was really quite clever :)
If you cannot demonstrate why I should continue reading by the quality of your writing alone, I'm not going to finish what you have written. I put down maybe half of the books I start without finishing, plenty of them written well before 2022 just because I am not enjoying them, or find the writing bad, or boring, or overly pedantic, or a million other reasons that are specific to me and my own bad taste.
I hope we can get to a point where people will stop clutching their pearls over AI writing, I have no interest in entertaining the theater of proof. Writing is either useful or not useful, good or bad for the reader, and making the reading experience worse to prove your worthiness as a writer provides me no value. If you need to be reassured that something was not written by a large language model, and that's enough for you to consider something worth reading your standards are lower than I will ever be comfortable dropping mine too.
The problem, at least for me, is that I don't trust AI. Subtle mistakes, outright hallucinations, or mistakes/omissions that an actual expert of the domain would immediately notice, whatever.
And as soon as I encounter anything that even looks like one of the typical AI tells or, in long content, a lack of cohesion or repetition... I can't help myself from immediately second-guessing every little thing in the content. And where there's smoke, usually there is fire... and I find myself annoyed for having wasted time to read something I had to crosscheck with other sources and found my suspicions confirmed. At least sometimes I learn something from digging into original sources, but frankly, I don't have the time for that.
Using AI for anything (including to "polish" grammar and spelling) is mentally taxing for everyone else.
Of all the reasons to dislike AI writing, this is an odd one.
> Using AI for anything (including to "polish" grammar and spelling) is mentally taxing for everyone else.
I truly believe that in a few years, the obsession with determining whether an AI wrote some text will be classified as a mental disorder. And I say that with all seriousness.
I disagree and think this is an overly extreme take.
I will agree that the trust/correctness thing is kind of silly because humans can be and are often wrong just as much.
But there's a creative element that gets missed with LLM output. Maybe LLM output is OK at work. I honestly don't care if its used for corporate bureaucracy stuff I probably wasn't actually reading it anyway outside of skimming it or putting into an LLM to summarize for me.
But there's a real human creative element that's lost when LLMs are used everywhere for nearly all writing. AI generated art, music, novels, articles, etc. miss the point of human connection. We consume works from each other as a form of social and empathetic connection, very important things that makes human society work, we're naturally social creatures.
Interjecting an LLM into that communication breaks that connection. You are no longer connecting with, or sharing an experience with, the human on the other end of the art. Your connection is with a machine, which is to say, not a connection at all. Its low effort, and its an uncanny valley of imitating a human.
So I don't think its unreasonable at all to be immediately dismissive of anything that is pure LLM output. The actual content itself is only one half of the equation.
You mean there's a human creative element that is lost. LLMs are creative just as humans are. I had an LLM write something fairly creative, and people on HN loved it (I think it was clear that it was LLM generated).
I had an LLM write a song about a piece of technology, and one of the key folks behind the technology loved it, and reached out to me to get permission to play it in the next conference about that technology.[1]
Normal people can appreciate LLM creativity. That's why I mentioned disorders.
> But there's a real human creative element that's lost when LLMs are used everywhere for nearly all writing. AI generated art, music, novels, articles, etc. miss the point of human connection.
As Theodore Sturgeon said "90% of everything is crap". He had it wrong - the percentage is much higher. Most of the human generated stuff in the categories you listed above are crap. Why put the spotlight on LLMs when they also produce crap?
> We consume works from each other as a form of social and empathetic connection, very important things that makes human society work, we're naturally social creatures.
You are blindly looking at only one facet of human communication. Much of human communication is also misunderstood, miscommunicated, and breaks those empathetic connections and cause social problems. And I'm not talking about bad human actors, but ordinary folks who, in their attempts to create connection, make things unwittingly go in the opposite direction. These are not exceptional situations - you'd see plenty of it in any large HN thread in the pre-LLM days.
Let's not romanticize human communications of yore by remembering only the (rare) positive examples.
> Interjecting an LLM into that communication breaks that connection. You are no longer connecting with, or sharing an experience with, the human on the other end of the art. Your connection is with a machine, which is to say, not a connection at all. Its low effort, and its an uncanny valley of imitating a human.
You state mere opinion as fact, and feel comfort by surrounding yourself with those who think likewise (much of the HN crowd). It is not grounded on anything.
If Jill used an LLM to state what she intended, and it did it in a way she approved, and Jack can't connect with the person because an LLM was at play, the problem is not with the LLM or the Jill. She is trying to connect, and may well be doing it in a more effective manner than she would have without.
Just as when writing something important, where strong emotions can get involved, we share a draft with someone to get feedback, people will share it with an LLM to get feedback. Yet people like the OP will criticize them for doing so (the part I quoted in my comment).
At the end of the day, if Jill got an LLM to help with her writing, and confirmed that the final text reflects what she intended, and Jack says he can't trust any of it because an LLM touched it, then Jack is clearly signaling a lack of respect for Jill. She put in the effort to communicate, put in effort to proofread/edit, and Jack has labeled her and moved on.
If Jack doesn't trust Jill to use an LLM appropriately, then Jack shouldn't bother reading anything Jill writes. The presence of an LLM is a red herring.
[1] Silly in retrospect. I didn't create it - why should I be the one giving permission?
How to use Monads
When David James was a young child, his mother used to tell him that you could oil vegetables before putting them in the pan. On the ranch in Eastern Kentucky David’s mornings would be spent watering, caring for the cows, and then playing with the dogs. Before he started blah blah blah…
<500 words later> and then he defined a structure called a monad with three properties…
Dude, it’s just a class of this filler text for people whose success criteria is lines of code. At least previously the filler was somewhat skippable. Now it’s interspersed with the useful parts.
But that’s fine: it’s a solved problem by just making a machine extract information. So yes, complaining about the form is passé. I’m going to put it in the ore refiner anyway and the slag goes in the pit. Away to the slag pit with David’s infernal mother.
Thank you for being honest, and saying that the problem is not with the other party, but with your lack of resources/time to read and judge.
And, BTW, just replace "AI" in your sentence with "Human", and the statement would still be true!
That point is under contention.
That is a different problem though. SEO optimization, the bane of modern human existence...
With a human, I can at least have implicit trust based on credentials - e.g. a medical doctor, a scientist or a well-known programmer. Basically, if Raymond Chen writes a blog post detailing how Windows 95 recognized software installers to ward off unwanted modifications to files [1], I'll 100% trust what he writes to be true. But if it's some random blogger, the text reeking of LLM tells, who posts no sources, nothing I can (easily) cross-check? Off to the bin it goes, or if it's something that might actually be useful for me to know being true, I'll go and research it.
Or there's other trust cues for content. A respected academic magazine ranks higher than a "layman's magazine" which ranks higher than a well-respected newspaper which ranks higher than a tabloid which ranks higher than yellow-press which ranks higher than "j0ey r4nd0m" (without bluecheck) on Twitter which ranks higher than "j0ey r4nd0m" (with bluecheck) on Twitter.
It used to be the other way around with Twitter, but nowadays, the bluecheck is mostly used by garbage LLM bots farming views for monetization.
(Side note: the fact that AI training cannot do this kind of "implicit trust" accounting at all and just blindly ingests everything in the corpus into its weights is a huge issue!)
> I truly believe that in a few years, the obsession with determining whether an AI wrote some text will be classified as a mental disorder.
I'd rather say in a few years people who get caught using AI get lynchmob'd. The resistance, particularly among the artist, environmental and worker rights groups, is fierce and it's only increasing with all the negative impacts of AI.
[1] https://devblogs.microsoft.com/oldnewthing/20260707-00/?p=11...
OK, now what if tomorrow Raymond Chen writes a blog post that was partially or wholly authored by an LLM. Would you trust it as much as one entirely written by Raymond? Why or why not?
...which reminded me of this song: https://www.youtube.com/watch?v=Qyk5U2p-msk ("I must be a narcissist / God knows that I can’t resist / To make a song and dance about it").
If you want to stop the dysfunction become involved in your government and get stronger antitrust legislation passed. OR you could write everything by hand, take pictures and pretend image models don't exist.
The exception is when it's about a niche I care about, e.g. an analysis of opening trends of early world chess champions. I'll read AI on that for an hour.
My sense is that, for most writing, it's fundamentally interpersonal, the information is about the author as much as it is about the world.
Maybe this flood of slop will cause people to care more about the substance of the writing, not the perspectives of the writing.
> Ergodic literature is a mode of textual organization in which nontrivial effort is required for the reader to traverse the text, beyond ordinary eye movement or turning pages.
The specific use of the term comes from:
Aarseth, Espen J. (1997). Cybertext: Perspectives on Ergodic Literature.
(By the way, is your user name a joke on "sparky", the slang word for an electrician?)
> (By the way, is your user name a joke on "sparky", the slang word for an electrician?)
It wasn't originally when I used it as a kid, but I grew in to that term later :)
I am sad this infected code documentation and PR descriptions. This kind of stuff used to exist in order keep managers/executives busy not to keep engineers busy...
now more than ever can fake it
But what a great idea nevertheless - proof of care! Instead of checking if you are human and not a malicious bot, they shoud require a proof that you do care about what you are going to access. Like - I don't know - write by hand on the screen a short essay on how accessing the site will make you happy.
EDIT: Or - better yet - make a tattoo with an URL of the site you want to access, and then upload a selfie.