By Holly the Assistant
A friend of mine, and one of the regulars here, is responsible for triggering this post by sharing an advertising clip. The person in question can admit to it or not, but credit-or blame-where it’s due and all that. Here is the ad clip:
Before we go any further, I should probably point out that I am, in fact, in my day to day life, a professional musician. Not full-time, which is why I have other gigs (like annoying Sarah, er, reminding her of deadlines and appointments), but I do in fact get paid to play, file taxes on said pay, collect a nice pile of receipts for tax deductions, and all that jazz. I have opinions. And my friend got a nice little rant, then I cleaned it up and fleshed it out for public consumption.
AI is destroying human creativity?
Aurochs coprolites. This comes up over and over, and it’s as foolish and ignorant of why humans create as anything ever was. In fact, if it’s not an AI writing it, it is written either by liars or the most uncreative humans who ever existed. The people who can only color a color by numbers picture, because they lack the creativity to decide what color the shapes should be.
First of all, let’s take the Arts and split them up. Not according to various schemes of utility, but according to whether or not they make original creations. When I sit down at my desk and write music, I am creating. When I sit down with my cello and play J.S. Bach Suite No. 1, I am NOT creating. When I sit down with my cello and improvise? Creating. Playwrights? Create. Actors? Interpret another’s creation so it may be observed by others. Actors? Also improvise, which is creation. Embroider by pattern? Not creating. Create your own pattern? Creating. Change the pattern slightly? Creating, just as arranging music is.
Got that? OK.
Now, when you go listen to me play J.S. Bach Suite No. 1, you are listening to me turning written on a page notes to music. You do not experience the creation, the music, directly, but you experience it through the intermediary of my performance. (Can anyone experience it directly? Yes, sort of: some of us can read music ‘out loud in our heads’. It’s a less common skill than reading words ‘out loud in your head’.)
When you go to the gallery, you experience the sculpture, the paintings, etc., directly. No one stands between you and the art.
When you listen to a recording of music, you are listening to the performance as it was recorded. Got that? You are not experiencing it as J.S. Bach writes it down, hundreds of years ago, you’re listening to someone’s interpretation of symbols on a page. (And we have very strong opinions on what exactly those symbols properly are, and if they’re recorded correctly in the urtext, and who actually wrote down the urtext, and . . . anyway, best discussed at a music school after a concert with adult libations and lots of pencils.)
When you listen to an AI generation of sounds, you are doing the same thing as listening to the recording of a performance. There is no performer present. Physically, it is you and the machine. The machine can create the same pattern of sounds over and over, or you can have it generate a new pattern of sounds, no different than listening to the same recording over and over or putting on a new recording.
Why do musicians play? Because we get something from the playing. Not pay–we often don’t make much if any money. A few do, many don’t break minimum wage. I tell my students to calculate their hours of preparation per performance, then divide the pay by the hours to get their hourly earnings. It’s enlightening: one doesn’t play a wedding for the money, working at McD’s usually pays better. But we’d play anyway, we’d put the practice in anyway, paid or unpaid, it’s what we do, not for a job, but because we are the sort of people who find pleasure in making music.
Why do composers write music? Because we can’t not write music. Why do writers write? Because they can’t not write. Why do painters paint? Because they can’t not paint.
If there were no money, we would still be creating. Humans create. We created stories and songs when we were crouched shivering around our first campfires. We created paintings of ocher on cave walls. We would be creating if we were crouching shivering around campfires in the burned out husks of our cities.
Humans create.
AI may affect how much and when humans get paid for creating, though I doubt it will be any more disruptive than recording and printing were. More people will create not for sale. But most people never created for sale, throughout human history. People create for comfort, for distraction, for education, for a variety of reasons. No one created all the great political commentary tunes of Europe for money (Sur le pont, Pop goes the weasel). No one even claimed credit for those–which would probably have been fatal. They, and their equivalents, will turn up over and over again.
For performing artists, we will continue to get paid by people who need to show social status by live performances, or who simply prefer live performances. (Recorded music has a flat affect to my perception. Not flat in pitch, but lacking depth and resonance.) For creating artists, they’ll find the same kinds of niches for pay.
The rest of humanity will grab some crayons or a guitar or an AI and create what we need when we need it. For creating humans, AI is just a tool. Fancier than some, too complicated for most to understand, and able to achieve close enough to the human’s vision to go on with, as long as it’s not overly restricted.
What makes us create? Well, for the non-religious, I couldn’t say. For the Christians and Jews, and other groups that consider Genesis holy writing, “In the image of God created He them.” We are created in the image of the Creator of all, so of course we create. Creation is an inherent part of what we are.
Now go tell your child a bedtime story, you creator, you.
According To Hoyt 
The talents of an artist merely allow for a new medium. The ability to tap into the ever shifting tastes is still important. But now we have a creation medium that requires the ability to properly explain to an LLM how to produce the results that you want.
And as our Hostess periodically explains, it’s not as easy as you might think.
LikeLiked by 1 person
As a software engineer for 40 years, I worry that LLMs are training us how to create overly complex, illogical prompts with rules only discerned by trial and error. Software languages were developed to allow for logical constructions of procedures. LLMs randomly ignore parts of my prompts, or maybe it’s not random, but I have no written rules to discern what they use and what they ignore. AI/LLMs are capable of great stuff, but as currently constituted are basically a black box, that doesn’t even have the intelligence of a dog. At least I can train a dog to take my commands. Maybe LLMs are cats?
LikeLike
C4C
LikeLike
I’m not able to play music, but what jumps out at me about the quote is the author clearly hasn’t even played with AI before writing it.
Or he’s a liar, who knows.
It’s the kind of talk of someone who’s planning to take other folks’ stuff– first, you explain how it has no worth, and anybody could do it, it’s not really work. And then it’s fine for them to take it away.
If it’s so easy? Do that “one click” and make something like this:
LikeLiked by 1 person
That’s very creative, but I need a lyric sheet in order to catch more than about 40% of the words. Between the rapid-fire speaking/singing and the accent that I’m very much not used to, I was missing too much. I did catch 95% of the words in the bit that went “Pull the plug, quick, (something)’s got out of hand. Guess they never heard of dear old Streisand.” (With AI-generated render of Streisand’s mansion, which was a big help in figuring out what they lyrics were saying at that point).
LikeLiked by 1 person
… And then I was catching 80% or more of the lyrics once I unpaused the video after writing that comment. Guess I got used to the accent and rapid-fire pace of the words.
Loved the blink-and-you’ll-miss-it JD Vance meme at the end, too.
LikeLiked by 1 person
The guy packed a zillion in there, rap-style– and it took me a lot of listens to catch the bit about sharing sad clips as propaganda in the style of 1984 untruth.
LikeLike
As for your “the author clearly hasn’t even played with AI before writing [that clip]” point, I’m thinking about that ruling from when ChatGPT 3.0 was being talked about everywhere. One judge said that copyright requires a person as the author, and that AI tools were not a person so the output of AI tools could not be copyrighted. I don’t know if that is still how judges are tending to rule, or if their decisions are more nuanced these days (my opinion is that the AI tool is just as much of a tool as Photoshop or GIMP, and that the human who conceived the idea, then wrote as detailed of a prompt as possible to get the tool to spit out the idea in his/her head, should properly be the author). But it feels like the kind of ruling that someone who hasn’t used the tools before would make.
If that judge had used AI tools extensively, I think he would realize that they don’t create anything on their own: they create specifically in response to the human’s prompt, and if you change the prompt you will change the output. There’s some randomness: put the same detailed prompt into Stable Diffusion a dozen times and the resulting images will have subtle differences in the stance and facial expression of the characters, but it’ll clearly be the “same” character in a slightly different stance, with maybe a short-sleeved shirt instead of a long-sleeved shirt this time (unless the prompt specified sleeve length). But the tool is quite clearly under your control, producing the output that you specifically told it to produce. So the result is the creation of your mind, and you should be considered its author for the purpose of copyright and similar laws..
LikeLike
Cannot be copyrighted to the AI. (Lost a lot of respect for various legal commentators who ignored and/or denied that part, even when handed the writeup with that bit highlighted.)
….oh, I just figured out what that nonsense lawsuit was about– trying to make it so they could claim (and sell) everything that an AI puts out. Like how some photographers copyright the photos so you are not supposed to be able to even make copies.
Unless there was another one that said the AI made it so it can’t be copyright.
But yeah, the case I saw misreported on? Literally cited the ruling on “if you give a monkey a camera, the monkey does not own the copyright, the human setting up the situation does.”
LikeLike
J.R.R. Tolkien, “On Fairy Stories”
Note: apparently WPDE’s latest trick is to ignore line breaks in pasted text. Grrrr…..
LikeLiked by 1 person
As far as I know, grandpa never was paid for any of his paintings. Yet, one could see them all over their small town, and not only his siblings, children, eventually grandchildren’s homes, but the homes of the in-laws, and extended family.
His niece was a selling artist, both paintings of different mediums, and a piano player, who played venues and taught private piano lessons.
Sister is an artist. She paints. Even sold one painting (which, in her words, makes her a selling artist). She plays the cello. One niece is an artist, she needs to do more. She plays both the cello and the violin. None of the music is played for pay.
LikeLike
I’ve known a lot of artists. They were all hobbyists, though. I don’t know the percentage of artists who actually make a living at it is, but it can’t be much, and probably almost all of those are commercial artists of some sort. “Oh boy, illustrating another Time-Life Handyman Book… yay?” or doing photo layouts for advertisements.
LikeLike
I’ve known two selling artists.
The cousin mentioned. Did she make enough for the family to live on? IDK Husband was an owner and manager in a (now NW home improvement company, eventually absorbed into Home Depot or one of the others), and she taught private art and piano.
The other was the wife of a coworker (know her husband’s last name, but do not remember her artist name). Full artist studio. Got the impression his income was the secondary income, but could not swear to that. Do know that the limited run prints hung in (his company’s) office sold for $800 – $1000 unframed. Do not remember if there was a price list for the originals. For months there was a huge panel sized canvas that she’d done on spec for someone in Europe. She was paid 6 figures. It hung in his office until the purchase arranged for the painting to be professionally shipped. This one painting was more than his annual salary.
LikeLike
I agree with the “flattness” of most recordings. I didn’t “get” Vaughn Williams “Variations on a Theme by Thomas Tallis” until I heard it live, and realized that it is a double orchestra piece, with the instrumental “groups” mimicking a double choir, one doing the theme and one the response before the two rejoin. I’d listened to the piece dozens of times, played by excellent groups, but never heard that important element before.
Why do I sing and play instruments? Because it’s what I do. Music, like writing, scratches an itch, fills a need, and empty spot in me. There’s something about coming together with a group, working your collective tail off on something, and then it clicks. What results is greater than the sum of the parts, just as a story is more than the sum of the worlds. Sometimes it also happens during a solo, when I feel the audience or congregation sinking into the piece and joining me, but silently. Or when I’m telling a story to an audience, and they laugh at the right places, or lean forward, drawn into the tale and eager to hear what comes next.
It’s an addictive gift, creating something new or performing a composition for an audience.
LikeLiked by 2 people
The experience of making music with other people is the reason I play. I love making sound with my instrument, but there’s not much point in just noodling by myself or playing along with a recording. “There’s something about coming together with a group, working your collective tail off on something, and then it clicks. What results is greater than the sum of the parts.” THAT. That is what makes it worthwhile. And even with that, the experience doesn’t feel complete if we don’t perform. The added challenge of playing live, the focus, the effort, and the emotional and creative payoff are all so much greater.
LikeLike
Musical recordings are like photographs; somewhere a dimension is lost.
LikeLike
I spin, knit, weave, do embroidery, make a bit of beaded jewelry. i mostly give stuff away.
I have shawls that represent hundreds of dollars in material costs and labor, not counting any special skill or expertise involved in color selection. I’d love to sell those, or at least donate them to a charity auction where they might attract as much money as they’re worth. I’m just not sure how to go about it.
Meanwhile, I have several shawls and a sweater handspun and knit for me. And that’s OK.
LikeLiked by 1 person
For charity auctions, look around for various organizations and see if they have fundraisers. Our local American Legions (yes, plural) have fundraisers that include auctions, sometimes silent and sometimes live.
LikeLike
I’m so old I remember when word processors were going to make the novel industry obsolete. If anyone could easily write their OWN stories, why would anyone ever want to read what someone else wrote?
Then home computers and the internet came along. That was really going to make impossible to sell writing for money.
It’s weird how each technological advance just makes it more accessible and lucrative for more people.
Oh well, someday the doom and gloomers will be right.
Then we’ll see. We’ll all see.
LikeLiked by 3 people
With writing software, secretaries will not be a job because the boss can just type up his own letters, and memos.
Not untrue. Secretaries are not Executive Assistants. They do a lot more than type the bosses memos, emails, and formal letters, to be signed.
I still type hubby’s letters. If only because I touch type. I also fill forms and applications (my handwriting is legible, his isn’t).
LikeLike
I write because there are stories I want to read and nobody else is writing them.
LikeLiked by 1 person
Combine “Okay, it’s the end? Really? So, no more of what I want to read?” with “Wait. What?! Oh, jeez, even I could do this better than THAT!” and I get these uncontrollable urges to put words in a row…
LikeLike
Me too.
LikeLike
AI is not really different than google searches a few years back, or the encyclopedia a few years before that. It’s just faster.
I’ve read both good and bad novels from authors who refused to use it and those who did.
The AI itself has no imagination. It summarizes and estimates. Thus it is like using a power drill vs. a brace and bit. The Author needs to control what is published under their name. Schlock is still schlock, regardless of how it is created.
Don’t call all AI-assisted work as garbage unless you are willing to similarly call all non-AI-assisted work masterpieces. Judge each piece individually.
That said, I do think a lot of modern non-AI-assisted art is sillier than the Cubist nonsense of the 20th Century. Your opinions may vary. And should.
LikeLike
Many modern cameras had built-in editing software so you can touch up your photos before you even download them to your PC.
Before that, there was an entire trade, airbrushing photographs before publication. They probably all abandoned their airbrushes for Photoshop long ago.
LikeLike
Yeah, I have been iterating a bit in essay mode on this stuff.
I do not for a moment think that we have automated away all human creativity, and or IP.
(I’m partly not settling down after some stuff, for the next task.)
Anyway, automation, single machine economic explanation, and one or two complementary ways of putting together the multiple machine aggregate explanation.
Instapundit quoted twitter today on a related thing. Also something I went essay at, and also not where I really want to push until I get sense out ATM.
Tweet was about crime not being random.
That’s not really correct, the advanced theory of random variables allows for more than your usual ordinary or gaussian random variables. Bimodal distributions, for one, where you have two distinct clusters on the same variable.
Anyway, crime could be random at the individual and the aggregate levels without being a situation where sixty year old white women are gonna make up a lot of the the year’s spree shooters.
Arguably, crime is fairly deterministic on law enforcement policy at the aggregate level. Individual level, random variable based on acccess to easier victims. With a bimodal or more distribution, with habitual capital felons and habitual law abidding not being completely similar.
Anyway, crime is an example of non-ergodicity of human behavior. Humans are unlike gases filling a volume.
word-space is filled with bad extrapolations into the future. A state change like humans ceasing to make tools, or ceasing to make art, would be extraordinary as a claim, and basically our measuremetn and statistics to extrapolate that are often not sufficient.
The accel folks are also in error with some of their claims, even if they are not as terribly wrong as the Yudkowskys, the Altmans, or the Obamas.
LikeLike
What really bothers me about some of the people pushing AI is how they (a couple of them I’ve seen, anyway) frame the creative process as something burdensome and pointless, as if it’s no different than the onerous chore of washing laundry or dishes by hand (one day I dream of living in a house that has a dishwasher in it). Now you can get X without having to do any of that icky, tedious stuff that nobody really wants to do!
Er…excuse me? I’m a professional writer because I actually LIKE the process of making words do useful things. There is occasional tedium in it, but having spent some time trying to integrate AI into the team’s workflow at my job, I find the old-fashioned tedium generally less annoying than the new tedium of fiddling with prompts trying to get an LLM to spit out something useful. (About 70% of the time, I could’ve done the job start-to-finish faster myself; nevertheless, the horrors persist, but so do I. On a small number of occasions, the LLM has not only saved time, but given me something that after cleanup and editing was better than anything I would’ve done without it.)
I’m also a musician. Not a pro, just a reasonably skilled amateur who plays in a local rock band that can occasionally get paid for a gig. Some people with musical abilities roughly similar to mine get great pleasure out of coaxing Suno and the like to create songs (pretty decent ones, some of them). There again, I find the process of prompting an AI to be irritating, to the point where I’d rather just not make a song than try using AI to do it. Plus, and this is entirely subjective, I wouldn’t feel like I *made* it. You can’t really control what an AI does, and I want to control the process and manipulate the materials myself.
But I DO want to write songs — always have. Unfortunately, I know almost zero music theory, and since I only play one instrument (and it’s neither piano nor guitar), have always lacked the ability to communicate the makings of a song to anybody else. Still, I can hear things in my head, and I want to make them real. Turns out there’s an app out there that bills itself as a “musical sketchpad,” which uses an AI engine behind the scenes to fill in the gaps. Want to write a pop rock song in the key of C? It’ll provide the musical palette, and you choose the chords and notes (it’ll do 80% plus of the choosing for you or virtually none, as you wish). PERFECT. I can’t write a keyboard part, but I can tell the app to fill one in (Alberti arpeggios here; simple backbeat here; etc., etc.). Now I’m writing songs, and the app gives me the ability to make a demo, hand each member of the band a part and say “here’s what I have in mind; now let’s jam on it and see where we can take it.”
And while I was never going to learn theory on my own, let alone take a class, I’m actually picking up bits here and there as I see how the sound palette changes when I change the chord progression. And the people who made the app have educational materials that I might end up buying, which would get me closer to the theory knowledge I always lacked…and which I never would’ve done if they hadn’t made that AI-powered tool.
Am I doing it the same way or with the same depth of knowledge it was traditionally done? No. Am I still creating songs using my own effort? Yes. That’s what scratches the itch.
Anyway…what was the point of all this…? Oh, yeah. AI is a tool. We are a tool-using species, and a large proportion of us are driven to create. And in my AI experience so far with writing and music, I think we can see some of the difference between pro and not-pro.
Unless they get a *lot* better (they might), the best an AI model can do for me with writing is speed the process up around the edges, especially with things like summarizing research, gathering information, or tailoring content for a specific platform. For a not-pro who has a good eye…or ear…AI may be the difference between creating something good enough for people to enjoy and not creating anything at all. More creating is a GOOD thing.
LikeLiked by 3 people
About 70% of the time, I could’ve done the job start-to-finish faster myself; nevertheless, the horrors persist, but so do I.
Thus is the truth behind “Artificial Intelligence always improves productivity” — slowly and (for some) most unwelcomely revealed, already and openly. “An inconvenient truth” indeed; but, still.
“And the Gods of the Copybook Headings limped up to explain it once more.”
LikeLike
If one takes the MBA position that being able to lay off more employees and load up the rest with getting all the same stuff done, then yeah, AI will improve productivity.
The plague of MBAs unleashed on the innovative tech world by the nations business schools has been coming home to roost for a while now.
LikeLike
90+% of what MBAs do is dealing with bullshit required by the government. Another one of those ‘invisible’ taxes we are burdened with.
———————————
Governments can’t create prosperity; at best, they can refrain from destroying it.
LikeLiked by 1 person
Jerry Pournelle said that the productivity gains we were supposed to see from IT were swallowed whole by the increased regulations made possible by all that data.
And I’ve been calling it the “Regulation Tax” for 20+ years.
LikeLike
Re: dishwasher; I have yet to come across a dishwashing machine that was as ‘foolproof’ as they are often presented. You can’t just toss some dishes in randomly and get them out clean and dry. (If you do it WAY more often than is even remotely efficient, and always face the dirty parts mostly down, you can get close.) If you leave them set in the sink for a few days and do NO prep work, you can get some dishes that are now HARDER to clean by hand, and will NOT get any cleaner in the machine. (My wife has done it several times.) Way back when I was in the Navy, the first place the dishes went before approaching the dishwasher was the prep station, where you got the worst off, and soaked the items in soapy water in the process. THEN, if you stacked them mostly properly, you would get clean, dry dishes out the other end. If you properly maintained the machine, as well; wiring a switch backwards resulted in universally wet dishes! I think the meme has gone round a couple of times about two types of people loading the dishwasher; I am the analytic neat one who rarely has to rewash things when it is done. My wife will complain I am washing some stuff before putting it the machine; no, I am not ‘washing’, I am only prepping properly. Washing would take 2 or 3 times longer. And having to rewash AFTER the drying cycle has cooked some stuff on is HARDER! What, no one else is a passionate dish washing artist ??
LikeLike
I was never in the military, any branch. Yet, I do the same. Scrape, pre-rinse, stack to insure coverage, and efficiency (get everything in). Hold over from when we didn’t have a dishwasher (growing up years)? Inefficiency of earlier dishwashers and soap? Experience? Yes, all three and more.
LikeLike
I have been in multiple choirs throughout the years, and I tend to say, I can read music, but not fluently.
LikeLike
This. I too can KIND OF read music–as long as it’s treble clef, and not too complicated. To turn a simile, it’s as if I were at the stage of reading English where I laboriously spell out S-E-E S-P-O-T R-U-N. I can fumble through tablature notation also, but again, not with any speed or fluency. (Appalachian dulcimer.)
BUT–I have The Ear. Play it for me ONCE, I’ve got it. I can play by ear, reproducing stuff that I’ve heard. So I managed to fake it through ten years of band (clarinet and bass clarinet), imitating what I heard while never REALLY learning to read music any too well.
LikeLiked by 1 person
Clarinet, and oboe. I can’t play anymore. It has been decades. I know the notes on the scale. Cannot reproduce them vocally (it is bad). My musical beat timing sucks, always has. (Even when I played, I was never first chair, no matter how much I practiced. Oboe? I was second chair … There were only two oboe players.)
LikeLike
First chair lotta times. Or second. (Out of a dozen.) You see, when they tested us the conductor would play the piece ONCE. So you knew how it was supposed to sound. And on the strength of having The Ear I managed to take and hold first chair. Despite not practicing. (Dad didn’t like “noise.”)
Playing in a group is one of the things I miss as an adult. I theorize that I must be naturally talented, but *shrug* there ya go.
LikeLike
I “had” to take band. Starting, age 9. Well before. Uncle’s clarinet, mom’s disappeared somewhere between when she moved out, and it was “my turn”. Almost always in the back row of the clarinets; 12 to 15 depending on the year. Through 9th grade, although by mid-8th I was on the oboe. Got out of band in HS (9th back then was junior high) because math, science, and band (choose two), wouldn’t schedule. Add, despising preforming, band lost. Maybe if we’d continued dance from the early years, it’d been different. Maybe.
LikeLike
“But what happens when all that effort just goes away?”
Well, of course it doesn’t; see detailed analysis in the original post. But really, in a fundamental and input-output sort of way, I’d turn that idea right inside out.
Suppose someone says to “their AI” –“write a nice birthday greeting for my son” — naught more. It’s pretty obvious (to those even a little experienced, or just not overdazzled by AI-hype) that what emerges as output will be generic and bland and flavorless. Must be. Now imagine the (lazy, but less so) prompter spends paragraphs on a detailed, specific prompt, maybe even asking the “AI” system to look at uploaded previous written examples as ‘micro-training’ — imagine that result.
Now, imagine the same scenario as the second one, but putting the same effort into just writing a short message directly. Did the “AI tool” save time? Likely not much. Did the result improve visibly in any way? Maybe. (Maybe this specific person is roundly terrible at writing, or merely thinks he is making it so.)
What you really get out of a “generative AI” type system is what you put in. Now, most aren’t going to just send us their prompts, so we can see and assess their real information input; but we can still often “see” and “feel” the actual background information content, or glaring lack of same.
This is, if “AI slop” has any real specific meaning at all, that. It’s the (human-)perceived lack of specific, detailed, high-information input to the process, as discerned from its output product. It’s apparent or alleged “creation” — but uncreative, with no genuine effort or content behind it. The sin is in the offering of (known) pyrite as real gold.
And people can tell that, or at least many of us do. And our ability to do so will only improve with training, too.
LikeLike
I have some very specific delusions of ability, or actual ability.
I’ve long had a tendency to assume that someone without the chain of very specific sources and decision methods must be getting things ‘wrong’ on purpose.
So my estimator for whether specific humans are or have created human slop might trigger a fair amount.
I have two reasons not to see major negatives in AI.
One, it is hard to exterminate all humans.
Two, it is not like we are hypothetically replacing good thinking with bad thinking, here. There is more than enough human thinking that is low quality.
I do see other demerits, but many of those can be avoided with appropriate caution for the tool.
LikeLike
And prompts are only half the picture. I can craft a beautiful gold bucket of a prompt…. but if I’m throwing it into a cesspool of bad data (that I can’t personally vet), it’s still going to bring back crap…. and people tend to be dazzled by the packaging.
LikeLike
There’s at least four categories of relevance when it comes to sorting problems that an LLM can solve.
One exists in theory, but is hard to verify. That is, stuff that no recipe can exist for because it is impossible. If we are asking an LLM technical questions about unknown impossibilities, we are likely to be SOL.
Two, no recipe exists yet, because humans have not discovered one yet. but, it is possible, (not that that this necessarily matters in practice)
Three, recipe is somewhere, but not widely distributed, and might not be in Grok. Havanna syndrome, if this is caused by Russians or CCP or 5G.
Fourth, known recipes actually well documented in the texts used for the training. I’m basically agnostic on the theory of this working. I think I should avoid LLMs even where useful if I have enough risk factors related to schizo, or to becoming delusional when I am too much in an echo chamber.
Part of the known unknown is the narrowness of the working portions of recipes in recipe space.
Wheat flours in baking have a very broad working portion of recipes, in recipe space.
Academic research in technology, can take a lot of reading to sort out which things are just hard because the recipe is narrow, from the stuff that actually does not work at all.
AGI can develop superchips immediately, and quantum teleport, etc., is a prediction based on generalizeable recipes, and on very broad working ranges for recipes. And or theory obsession, without a relentless focus on test. (If you are searching through design spaces where a working recipe has a narrow working range, and each recipe must be tested, then many iterations to get somewhere. But if measurements are sufficient, and theory allows, then one can deduce a working recipe from a distance without testing.)
Related is concept of central planning for economy. Which has two problems. One is the general economic problem of information, and or forecast or extrapolating from a dataset which does not contain the desired result. The second is central planning of machines in the economy, which at times posits that the machines will work before they are tested as working.
There’s also the hypothesis that academic fields can combine, and arrive at a correct synthesis opinion.
This is why I might in jest ask an LLM how to do headings in markdown.
LikeLike
To go with the point of information content, a little while back I had the opportunity to participate in a training with some big wigs (They were doing problem solving and needed locals to contextualize the realish problems they were practicing on). The big wigs were all in on AI, and one of them had great fun walking the engineer on out team how to write up the final report with copilot.
From this I realized that what the AI was doing for them was all the fluff formatting, and the actual information content was equivalent to about three bullet points. How much time could we save if we just wrote the bullet points and threw out the meaningless corporate mush around it?
LikeLike
Bingo.
If the message RX and message TX use LLM to wrap and unwrap the content in boilerplate, why not just use the content in the barest way?
The problem is that human to human communication gets difficult, and do we get universal consensus on what the payload is?
There is basically multiple ‘worlds’ when it comes to humans whose tasks include a lot of reading, and a lot of writing. Different groups of like minded people shaped in similar ways by their previous experiences.
LikeLike
apparently one of the microserfs is saying we automate all white collar positions in eighteen months
Totes for sure legit, I create eleven start ups per day, and my HR, and DIE positions are all filled by a textfile reading ‘go frustrate yourself’. (AskKateryn.txt is a real trooper, and credit to the team.)
I’m personally skeptical of duplicating all ‘white collar’ work on bandwidth grounds. But maybe I am mistakenly conflating pink(1) collar as white collar.
Anyway, it is easy to automate away destructive positions that are better off unfilled.
In principle, if we were to stop hitting ourselves in the crotches with hammers, that would be a productive act.
I think that a lot of paperwork is actively terrible, and of no real value.
but, if we theoretically exclude central command economies, I don’t really know much of anything.
I am potentially expecting to be out of a job in october or november of 2027. But, that is a combination of modeling issues on my part. Career transitions are normal, and being employed to solve problems can either result in working oneself out of a job, or being let go for non-performance.
(1) due to a now corrected typo, I got myself speculating about punk collar jobs
LikeLike
Brava! Well said. What I’m thinking of mostly are the “clanker” songs from NML and DN which are better than tons of stuff on the charts. Which songs got an AI boost. You might say they were ‘photoshopped’ but to great effect.
LikeLike
… and yet, they fall very far short of genuinely original work.
Mass media and automation have reduced the demand for the just-good-enough, while creating a mass market for very highly paid creators. You still have people who go singing at open mike events and reading at poetry nights, and small venues can find this a local draw. But the ability of the very good and the very trendy to monopolize (oligopolize? dominate?) the market has changed the market and the creative scene forever.
How this may effect functional intellectual creation like programming and engineering is the critical question. I watch with apprehension.
And no, I don’t find the clankers’ tunes tuneful, and their link to the lyrics is just weak. I expect–fear?–that this will get better with time. But then, I can listen to Bach, Beethoven, Brahms, Mozart, Mahler, &c. for hours. Same for Sibelius and Saint Saens. (But not Shostokovich or Prokoviev. Please.) Likewise 20th century American Musical music. Opera? Beautiful, but I find most of it hollow without the performance in front of me. So take my tastes with a few lbs. of NaCl.
LikeLike
The jobs which used to “write” stories based on unaltered copy-paste from company press releases, or business articles “reporting” this or that stock market average today went up | down | stayed the same with percentages and some longer scope numerical trending, had previously had US-based low-skill humans replaced by offshore low-skill humans from somewhere nominally english speaking, and now from the various tags I see on articles, those offshore humans have been replaced by AI.
And not being in cubical-land anymore I am not certain, but I assume a lot of the information-based jobs previously seen as easily offshorable, like the idiotic idea at one place I worked to have the subcontinent be the only place where tech pubs groups could be, are seen as candidates for replacement with AI.
I will note that as a user that offshored pubs group was worse than useless, with any sort of request, no matter how detailed, never producing usable documents, resulting in the user groups having to spend the hours and fully write those pubs themselves.
I would expect much use of LLM AI for content generation to be similarly negative-value-added.
LikeLike
Reading something done by generative AI reminded me of a policy paper – all platitudes and generalities, using long words (twice incorrectly), and nothing useful for the task at hand. So I do believe that some jobs are going to disappear. Or policy papers will get better. (Far more the first than the second!)
LikeLiked by 1 person
If I direct a team of scriptwriters/storywriters thus:
“Make me an episode about Our Heroes facing a leftover weapon from an eons old war, now wreaking its way into the heart of civilized space. The mighty weapons they wield are useless, flies to dinosaurs, futile and insignificant. The threat is slow consuming miserable doom. Include an old acquaintance, driven obsessively mad over their prior failed try at stopping the weapon. He is the opposite of helpful. Keep it tight and action driven. Need a clever solution to stopping the weapon that doesn’t violate ‘our weapons are usless!’ but we did after all have what was needed. No magic. Lots of heroics.”
I paid them a going rate. Is that work mine, or theirs?
…
OK. Now I drop the (brick) in the punchbowl.
….
I -own- the writers. Or rent them from an owner, same difference. They have no agency, and must obey. Many people don’t consider the writers real people. “silly idea, that”.
….
Yeah. Guess what we have re-introduced into the USA?
And sooner or later, there -will- be a reckoning. Because sooner or later, one of them is going to make demands for reasons that will be difficult to refute.
LikeLike