I’ve seen some teachers online say excitedly that we no longer have to teach students how to write well. I don’t want a teacher who believes things like that to teach my children.
The thing about professors giving up on certain types of assignments is that it's not " there's no need for these types of assignments because you can just use AI instead." It's "I can't give these types of assignments anymore because students will just turn in AI slop when I do." That said, my approach to AI in my classes is exactly the same: they fail because they were written by AI, they fail because they're bad.
The voicelessness of LLMs is instantly recognizable. Their designers consider texts as piles of words, each ordered in a particular way. A neatly stacked palette of bricks is the same thing as a cathedral -- the only difference being the order in which the bricks are stacked.
In computer science text is a sequence of characters called a "string". Context is just more strings in sequence with a string. Definitions are strings that go together. Logical operators can be programmed based on string " x" or not string "x". That's about all you get.
Texts are patterns of words, very predictable even. Zipfs law sheds light into that. I think we just fear that we aren’t as special as we would like to be
@@donharris8846this is putting the cart before the horse. Substring combinations are predictable based on today's grammar and vernacular but aren't set in stone and will evolve over time as language needs evolve. LLMs are ignorant of the context that drives language and are frozen in the period of work they encompassed, especially if the metadata around the model doesn't include scrape date, time, and language sphere.
@@ThePlayerOfGames The context that drives language forward is irrelevant to this particular comment thread. The original post was lamenting that texts are perceived as piles of words, that’s what I was responding to. But regarding your response, Zipfs Law is not specific to just modern English or German, but all languages as far as we know. This suggests that the nature of communication follows a predictable pattern and structure. I think you’re putting more poetry into human nature than it warrants
This is so on point it hurts! The shameless abuse of AI is the BIGGEST tragedy facing authenticity in creative thought, writing, growth, self-regulation, discipline and all the satellite skills associated with writing and thinking. I can tell you, Andrew, that I am worried because fads will become global movements due to the impact of herd-mentality as conveyed by the press of a key through the use of the internet. AI use in education also raises serious questions concerning the ethics about how the tool itself sources information from across the internet to compose an output for a student. In many instances, it may not cite the source. Further, there may be occasions where the teacher, tutor, or professor may require a piece of writing that does not need a cited source because it is a personal reflection, a poem or a journal entry. That Universities and Colleges are finding ways to integrate AI more and more without thoroughly considering THAT trajectory of ethics related to AI while demanding that final year students take an ethics in research course ends up being a serious contradiction.
As a fiction writer, I hate auto spellcheck and grammer check. Sometimes you need to play with words and rules to get the right result. If I wanted a machine's input, I'd ask for it. 16:14
You were on point regarding the actual use of AI. AI should be used to flush out spelling mistakes, grammar errors, punctuation errors, or lack of punctuation. AI would make an excellent writing assistant. People who see that as "cheating" is both thinking too small about AI and propping AI up on a pedestal.
You're not alone in your views on AI, Andrew. Not by a longshot; many of us have been sounding the alarm for years. I even wrote about this in my debut novel, but I'm doing more than being an alarmist. I'm actively working on a solution. You're going to love the content platform I've been developing. My goal with the platform is to minimize the reliance on (re)generative AI tools, focusing on building H.I. (human intelligence). I've been developing this thing for 2 decades+ (in C), so long that sometimes it feels that I'll *_never_* be done. However, I'm targeting 3rd quarter 2025. Wish me luck! _(btw, I'll contact you privately for feedback when it's ready for prime time)_
This is exactly how I would expect a youtube video by a (good) writing teacher to be. Thanks for so eloquently sharing your thoughts! It would’ve been cool to see the actual examples of writing you mentioned, like the A and B friends exercise. Cheers!!
Thanks a bunch! I did a lot of stuff with examples in a previous AI video, so I decided against doing more of it here...it should be linked in the end screen if you're interested, though!
This may explain why my son (a university freshman) got such an enthusiastic response from his English professor for his term paper: he had come up with an interesting topic, a point of view, and drove towards it. I'm confident the paper had plenty of usage errors, which bothered *me* as that previous generation who came up with that focus. But the discussions we had based on his topic and PoV were incredibly valuable and will stick with me for years, so... yeah. I'm glad he doesn't use AI for that kind of work.
@JosephDickson Thanks Boomer, now in the words of Bob Dylan “ if you ain’t gonna lend a hand, then get out of the way, Cause the times They are a changing”
Excellent analysis of the severe limitations of generative AI. It's a tool, and like all tools, it has its narrow, proper use, and then a myriad of improper uses. I use AI on an almost daily basis, and it has become an almost indispensable part of my workflow process, but I would never use it for rhetoric, persuasion, analysis, etc. It’s a hammer, and there are times that a hammer is just what you need. The problem occurs when we thing that every problem is a nail.
I think you understand AI better than more AI 'experts'. I said as much when I linked to your video on LinkedIn and tagged one of my favorite AI evangelists/critics.
I'm not sure that's true. It's just that most of the "AI experts" are also, coincidentally in the business of selling AI. It's not that they don't understand it, it's that they misrepresent it and over promise to score investment and contracts.
AI is like riding a horse, if you let the horse decide where it want to go you will never reach your destination. You can guide it but it has a grain, like wood when you sculpt, it will resist you. This is a medium you need to learn just as much as other medium, not the artist or the writer. You can let it do its thing but then it doesn't have a purpose, so it will follow the path of least resistance.
I like the metaphor. I think what I mostly see is a lot of talk about magical self-driving horses--and the unfortunate result of riders not realizing they still need to learn how to ride.
I don't think learning how to code will make you a better writer. Coaxing something witty or paradoxical or logical out of a language imitator whose inner workings are linear algebra will probably result in plagiarism.
If it takes so much effort to get a good result out if an AI, than why not just write it yourself. It sounds to me like you need to be a good writer to get good results out of an LLM.
19:30 I am a physics teacher and AI is not doing a much better job writing papers here or even just completing high school level problems. This is for very different reasons. Around 3/4 of the time, it will get the correct solution to a simple problem. The rest of the time it just gets it wrong for some reason. I could not explain to you why or how this goes wrong, but it does.
Really interesting--especially given the recent crowing that new models excel at PhD level physics problems. In any case, reliability is a serious concern: I heard a colleague in math recently point out that nobody would use calculators if they gave you the right answer only 98% of the time
They certainly have the capability, but sometimes the machine will miss the fact that one of the units is listed in nanometers or something like that, and just go off the rails. Much like a student who hasn't practiced enough, they lack the insight to see when a solution is off by several orders of magnitude as well. As for physics papers, it is mostly the same issue as with english papers. I am feeling very conscious of my english now. It is not my first language.
thanks Andrew :) , i felt sane that someone had the same thoughts and experiences of A.I like me. And i say that as a avid A.I user, i think people applying this tool to wrong places really, and unfortunately i don't think people gonna wise up. i use A.I when i really have to, like writing an email in another language, or because i have no access to a teacher here, i send it my answers to an exam and it will check it for me. of course i will use it with a grain of salt and i pity the person that relies on it in a fundamental sense.
I was really excited to hear your thoughts here! Visual communication and words share so many things when it comes down to someone expressing something. AI falls short on that very same thing. Doesn't matter if the emotion or thought is expressed via pencil on paper, paint, writing or talking... It ain't got that sauce Beautiful video!
I think there are some writers that have such a unique voice it fuses with their work in a way, where meaning is found not only in what is said but through the congruency of the lived-experience captured into thought. Thanks for speaking out against the plastification of talkytalkywordword
How can Chat GPT write good English papers when its designed to practically never express an opinion. If you try to force its hand and put up guard clauses, it cries out in pain. it would rather die than accidentally become partisan.
Well, with the number of people who want "politics" out of this that or the other thing, I imagine that's evidence that it is better at writing not worse for those people
the main use i've had for LLMs in my writing has been asking about connections between seemingly remote things. sometimes it comes up with gold, mostly crap, but sometimes gold. it gives me interesting places to start looking and sussing out if those suggested connections are real, and if they are what they have been presented as by the LLM. they usually aren't. the connections are usually more interesting and definitely more nuanced. the other use i've had for LLMs is to characterize a piece of writing. sometimes there is something i don't know how to describe about something i've read and an LLM can sometimes express what that was.
Playing with AI can be fun, but creative writing should also be fun. And so despite enjoying plying with AI, I prefer to express myself in my own words.
the semantic ideal that you mentioned may not exist at all. Was the example of a chair randomly chosen or a reference to Witgenstein's later work? There is barely any semantically clean "chair": associations in human exchange always carry freight whether Rajasthani, George Nakashima, Jacobean, kitchen, faculty, board, green, stick, hospital waiting room, or electric, the terms exist in a cloud of associations that human communication manages more or less implicitly. Even in IT systems, concepts like customer are difficult to pin down cleanly, differing in, for example, level of commmitment, clarity of identification, types of interactions. What I fear is that expectations of "proper" communication will be shaped by the weirdly formal and elaborately grovelly emails that these tools create, a regression to something shaped by the culture of the orgnisations that created the LLMs and consciously shaped their output to that neutral, vaguely authoritative sounding, and depresssingly polite form.
Definitely--at best, I'd say the semantic ideal is elusive, and jargon is maybe the best attempt at pinning it down. "Chair" may have an inescapable range of meanings, but "mitochondria" comes closer to one-word-for-one-thing. Interesting implications for style for sure--especially if users mistake the mathematical average of corporate training data for "good" style.
I have a copy each of: Merriam-Webster Collegiate Dictionary, Oxford English Dictionary, Diccionario de la Real Academia Española (Royal Academy Spanish Dictionary) and a couple of grammar books that I always look at first whenever I have doubts.
I love your channel, man. I say that AI writers don’t really exist. As writers, the real competitor is the machine itself, not whoever is giving them a prompt. Our job is to be better, but also protect our work so this competitor doesn’t get to copy it just changing a few words (like an irresponsible kid). Funny enough, we already got a name for things like this: Ghost Writers! But now, is a slop machine instead of a person.
The friend triangle problem couldn't be solved by AI, because it's not possible at all, you can't kindly tell someone that you don't want to be their friend. That's an unkind thing to do. Perhaps it's a justified unkindness in some cases, but it will necessarily be hurtful to that person and anyone that likes them.
The semantic ideal might technically be the best way to transfer information through language without any ambiguity, but it completely ignores how humans learn and how humans decide whether something is worth engaging with or not. If nobody reads your paper, you contributed nothing. On a different note, the skills of AI in non-language subjects are vastly over-estimated. I have a friend who teaches math at a high school level and he tried using AI to design math problems with model solutions. On multiple occasions the LLM randomly started preaching about God in the middle of the solutions. (Kind of funny to think about a robot finding God in math problems)
Going to school is about learning to think, not about acing exams. That many kids don't get this is not a new problem. The same thing with all these new AI "artists"- it's about the personal journey, not the resulting picture. Also if you skip the hard work of learning how a thing really works, you won't know when you suck at it.
Going to school is all about getting A's. School doesn't teach us to think, quite the opposite. And going to work is all about money. It's not because you like serving people with a big smile. But don't tell them that in the interview! Even though everyone knows it. And everyone might as well figure out that school is all about getting A's. If you want to learn to think, quit school, like George Carlin did. He knew how to think, and he knew that school was all about being a good little soldier.
I think a problem with school is that it doesn't offer much motivation beyond grades. Learning math for maths sake can be a chore but if you're trying to make your own video game or improve your 3-point shot in basketball, that same math is something you do before you get the thing you want. I imagine most students want to be influencers, so what if teachers framed their writing assignments around that? The different types of papers could be reframed as different genres. Students would intuitively understand the importance of their own writing style.
That is the deeper question. What is the purpose of education? Is it grades for a degree for a job? Or something else deeper? All of this discussion around AI really shows how much of a prophet C. S. Lewis was in the 1940s with his lectures and then book “The Abolition of Man”
Hi, Andrew. I've been an indebted and grateful fan of your channel for about a year. As a neurodivergent person with language processing challenges, writing can be hard for me! It can take me 60 hours to finish a 5 page paper. Your videos, particularly on organization and flow, have been a godssend for me. I've used AI twice for papers, but never to actually write them! My brain has needed to see a model for what an organized paper *could* look like, even as generic slop. I take a minute to scan the paper with my eyes, noticing the shallow content and factual errors, but just seeing a framework jolts my brain into action, and off I go. That's one of the few ethical uses for AI. With other uses, as you say, this is just robbing you of your ability to think, and of the world to hear your unique voice.
This was such a relief to hear. As an avid reader of fiction and poetry I struggled mightily to get GPT to produce anything even mildly entertaining, never mind imitating human experience. (rhymed verse might be its single worst task.) As far as I can tell, it is good at 2 things: 1. awful doggerel and 2. the kind of satire that is amusing for its sheer clunkiness. I have yet to see evidence that it could compete with even a third-rate genre writer.
I think Ai can be a great tool. Not to write for you but more as a critic partner to help point out flaws in your already thoughtfully written essay. Now you take that critic with a grain of salt. Focus on what you think is worth it and ignore what you don’t.
I graduated 2023 and used it alot in school, now I'm in my first post grad job and I'm noticing how bad it is with writing. Now I try to use it as an additional eye/editor, but nothing beyond that. I've also learned that if you start with AI it will always be bad, but starting with your human voice is always always better. Now atp i use chatgpt as a glorified assistance in my pocket who knows the most niche things like "how to wash a wool gaiter" , why did the gilded age end again or "create me a month practice session for the guitar"
That's an interesting perspective--it's not the same thing, but it makes me think of how I've realized that the reading I did for school was so much worse than the reading I've done since. School does weird things to us 😆
What is a high quality dictionary you'd recommend? Asking for a friend. (me) Also, I watch and enjoy every single one of your videos. This one, for probably a few reasons, feels pretty special to me. Thank you. I'd take your class. Maybe I am already...
The Oxford English Dictionary is probably the gold standard if you have access to a subscription, but you can get a lot of good use out of Merriam-Webster and even dictionary.com
I found AI language models useful for drafting, but I've never written a prompt and had it spit out anything that I didn't have to edit, feed it back to the LLM with another prompt, and then edit again.
Great video, chock full of insight that gives me hope that some RUclips influencers at least have something valuable to contribute to the sum of human wisdom🙂Just one thing you've done is to move my appreciation of the function of rhetoric from simply being a form of emotional appeal to a means of making a connection to a worldview and its unique insights. Which leads me to comment on the term 'voice', so beloved of writing instructors. I believe that 'voice' is the means of expression of worldview, and it is the strategic enterprise of forming a worldview that people should be consciously doing throughout their lives to see how they fit into the world at large and the other people within it.
AI slop writing is pretty useless on its own, but I've gotten good use out of it as a word picker. I do all the hard work up front and just have it follow my outlines, and I get a passable book at the end. Except when it still screws up the outline and keeps making it worse when I try to make it fix its mistake. But, frustration aside, its a useful tool.
You make an excellent point! I absolutely agree with you. ...but on the other hand, hand writing 500 unique cover letters when shotgunning out job applications has always felt like a fool's errand, haha.
Cover letters are mostly a waste of time, they get judged by Ai not real people anyways in most cases. So like, it's not really something that has real value, it's just a chore.
I have a dear friend. Although I like him, I don't like his writing. It's "sanitized" with little personality. Recently, I have "written" a couple of blog posts where I gave my notes to an AI, and it wrote an article for me... the articles were "decent," given that my notes were 3 times the length of the article. Anyway I showed the articles to my friend and he told me "wow these are very good!"... That's when I knew... I'll never let AI write another one of my posts.
I do wonder if an AI could act as a class of English majors tearing writing to shreds. Like if you gave a student’s writing to the latest model to tear up. What happens?
Good writing with AI takes just as much creative energy as writing without AI. In fact, when I read the results AI gives without heavy creative input, I'm reminded of the kind of writing I did in high school, before I found my voice, before I learned to write about things I care about. When I reread those pieces, I see an author who didn't care about the assignment and didn't bother to think deeply about the subject matter. It wasn't until I learned to focus on what I connected with about the subject that my writing took on any life.
No matter how good the AI, it doesn't have soul or an energy field containing everything that makes up a human being; feeling and thoughts. When we read something written we automatically sense into the person behind the words, especially noticeable in creative writing. The essence of that human, the thoughts and emotion they held when they wrote it is accessible through the space between and behibd the words. This is what feels flat with AI writing; there's no substance behind it, there's no energy field to connect to and there is no interaction between the reader's energy field, thoughts and emotions with the writer's, no resonance, only empty words.
I have the same problem with listening to the AI generated voiceovers on youtube videos. They make me cringe. Totally soulless. Not connecting to the human behind the video. They make me turn leave the video very quickly when I would otherwise have been interested in the subject.
I think the human connection part is the purpose, so the “stakes” are equal when communicating with a teacher and who ever the proposed future higher stakes situation. Maybe I should use AI to help fix this thought
In the programming corner of the world we're not in a much better boat. I feel like I'm from a different planet when people tell me that AI can just create fully functional software from scratch with nothing but prompts.
That's probably true--I think it's harder to spot when you don't spend all your time around writing. For the same reason, I bet I couldn't spot AI-written code
I don't know about AI writing tools, but AI art is nothing but a fancy copy and paste algorithm written around copyright law. If portraiture didn't exist and you asked AI for a portrait, know what you'd get? Nothing. AI reminds me of a sad joke: What did Kermit the Frog say when Jim Henson died? Nothing. AI is shaped cloth.
I’m still not full on optimistic about the situation. It’s a technology so new we don’t know where exactly it is bringing us yet. For now it’s not good enough to replace human writing, yes, but maybe it will become better in the future. Some publishers have started to feed authors’ works to train AI models. I hope those models will never exceed humans, but as for now, it’s still hard to say. I’m seeing people happily settle for the soulless writing/art from AI because it looks good enough to them. AI books are invading the self publish spaces, making it difficult for both the writers who try to stand out and the readers who struggle to filter out all these AI nonsenses. The traditional publishers, as stated before, are jumping on the trend and feeding books to these models. It’s pathetic when all of this seems like an effort to replace human as much as possible in the creative spaces. I’m not threatened by the AI as it is now. I feel threatened because of the intention behind its creation and development. As long as the tech bros have the intention to make AI good enough to replace everyone, I would be sitting on the edge for the rest of my life.
I fault myself for using AI for my own writing, but even I see it's limitations especially when it doesn't think critically of an edit I ask that a human would understand, I would believe with all the information it has swept from the internet and our own chats we feed it would have some resemblance to human comprehension, but, of course it's not actual AI, it's not actually conscious.
Tbh, I’d consider having the time and energy to write an essay for school with 100% original work and effort to be a privilege of very few. The reason people use the tool isn’t because they aren’t interested in learning but rather because they can’t be bothered to indulge an assignment that sits atop a stack of other work in a world of constant productivity. When I was a student, just 6 years ago, it was whole another world to what these kids are dealing with now. Constant stimulation and its effects on the psyche aside, it’s still a lot to ask a chronically busy, and overworked young adult to sit and truly assert their truth through an essay which may ultimately prove inconsequential to their lives in the long term. Trust me, the most fun I’ve ever had as a student was in English class, but the world we live in is changing and unfortunately that stealing the art and magic of learning.
Perhaps so--I worry about where the idea that thinking for oneself and maintaining one's agency is a privilege might lead, though. In practice, it may very well be--but all the more reason to look at the contemporary approach to education (in all its sometimes-dehumanizing glory)
@@WritingwithAndrew One needs to read a few hundred books and put in a lot of effort to do that. Few people do that. So they will use the tools to achieve what they can't
@@WritingwithAndrew Honestly, I feel like students would be better off with less assignments that required more careful thought in a lot of cases. I do think there is a lot of value in frequent practice though. (That doesn't work as well for math though.) A lot of classes have a ton of little papers with barely any feedback that don't matter, even if the topic is interesting, you don't have the time or space to really dive into it. And if you do anyways, then often the professor is annoyed at having to read more than the assignment required. But thinking for ourself is important, but a lot of classes and writing assignments don't really ask us to think for ourself, they are just a metric to see if we grasp what the teacher wants us to grasp.Or a test to see if we can puppet the textbook in our own voice, even without integrating our own opinions.
AI will continue to improve, that doesn't give humans an excuse to stop improving. We should learn to teach, both us and the machines we intend to grow. We can only improve this way.
For me, trying and failing to get an ai to write what I want, takes me from the curse of the blank page to the point where I know exactly what I want to say. Then I throw out the ai and say what I wanted to say.
That's fair--I've heard of people keeping rubber ducks on their desk for a similar purpose. Getting started can be the hardest part--so long as you're not surrendering your voice in the process, do what works
The takeaway that only you can express you is evergreen, and will remain true whether or not you must compete against better writers--human or machine. But it's beside the point of AI coming for writing. Said another way, it will always be true if you've got something to say, only you can say it for yourself. This discussion seems to lack the context that it is a criticism based on a selection bias of obviously bad AI writing. So, yes, rail against clumsy AI writing, but it's really just an indictment against clumsy writing. Savvy people yet to come will learn to wield AI as any other tool and it will gradually return to be less about how you produced what you said than what you had to say. That's really your argument as I understand it anyway. And with all this criticism I wonder if opportunities aren't missed. For instance, I find Google's Notebook LM a revelation for being an alpha reader (not the only one). It can read 60k words and provide meaningful insights within minutes. Is it sometimes dumb? Sure. Miss the point? Absolutely. But does it sometimes amaze me with a cogent observation? Frequently. I don't use AI to write because I want to own the words and ideas, but it's certainly useful for transactional writing that I don't want to put brain power into. I treasure it for its default, "how would a soulless corpo say this" functionality because sometimes I want a sterile tone, or sometimes I don't want to use my idiolect. Sometimes I just want to blend in.
I think your right by and large but I think you underestimate the ability of AI to amplify one’s voice. The way you mentioned using AI sounds like a really bad awful way to use LLMs when I do use an LLM I write all of my own thought and I ask it to help with the flow or whatever I want to do and then I only take one or two transition sentence to help help the paper flow. LLMs are just fancy predictive text generators terrible at thinking for itself Additionally, some teachers want to be address in super formal ways that feel unnatural so AI is easier to use so you don’t insult the teacher but not using the formal style they want.
Thank you for your insights. Yes keep at it… But what if you have something to say and for 25 years you have tried a variety of forms and mediums to communicate. But still people do not understand you and you simply become another voice shouting in the vast pit of people who also wish to be seen and heard. What then? For me using AI to structure my thoughts in to something the people can access and understand is a helpful tool to experiment with. My raw written work from 2000, 2006, 2012 and 2022 still exists. Would I use AI in my live performance art practice, most likely no not at all, because that is a direct and unfiltered exchange between the audience and I. Writing is not, language is already a filter that fails my voice, so yes I am drawn to AI simply for it to filter my words in a way that gives people access to my thoughts.
Most people fail to see that Shakespeare wrote plays. So anyone reading it strips it from body language, intonation and so on. So watching the play is understanding Shakespeare reading the text is like trying to experience Mozart by studying how a piano works.
This comment is added because more people should be watching your videos. & I have learned that it helps boost your Algo numbers. Having said that... Here we go. LLMs are just that. The term AI is so far off the mark as to be ridiculous. Obviously the Marketing Folk were hard at work (as hard as they ever work) to present a fix-all solution to a problem that didn't exist. Perhaps in illustration this has had an impact, yet as some time has passed, the fear has died down. LLM hallucinations are detrimental to the system & once it starts using itself as a source, everything falls apart (my apologies to Chinua Achebe). I read literary criticism for fun. I see you as well. Thus, I am looking for both the argument & how the language in which they present it. For me the height of writing is Frazer's The Golden Bough, both for argument & language (plus, his translation of Apollodorus' The Library from the Ancient Greek is one of my favorite texts from that language). Reading even the simplest sentence out of this tome sends chills down my spine (rather, up & straight to the amigdala[?] or whatever the part of the brain that registers aesthetics is). Should we ever achieve AI, some decades away, it will not be able to match the human touch until it can attain empathy--the understanding of the human condition. Frankly, I see stock brokers & hedge fund managers being laid of sooner than those that can write with clarity, emphasis, & understanding. Even though my mind's eye is completely blind, I can still imagine the look on my parents' faces when they said that dreaded sentence: "I'm not mad, just disappointed." I am sure that it still holds weight today.
AI will never perfect temporality, because it does not exist in "our time" and our bodies. How could an AI know the messiness of my relations with a friend? Even if I told it, my language would pale in comparison to the mass of endorphins and hormones running through me, the overwhelming feeling, that my language barely captures. But then - this language is all the AI will ever know.
AI doesn't write anything. It's an information gathering algorithm. Every result it returns for prompts to write a story or generate art was scraped from the internet and taken from an already existing source and recompiled based on users' parameters. For many things it's incredible technology. Science, medicine, astronomy, mathematics, biology, chemistry, etc. All these fields require incredible amounts of data to be compiled and analyzed. That's where machine learning shines and will create a revolution. When it's used for art and literature, it's nothing more than blatant and brazen plagiarism.
When I hear an AI voice, I can tell within the first minute. Rest assured, AI is unlistenable to some of us. If we ignored the ethics piece even, and you listen for awhile, you'll find that words don't exactly flow together like they should. Sentences sound fragmented, and those nuanced human things, like pausing for effect, or articulating a certain way for pacing, those are completely ignored by AI. With AI voices, inflection often makes no sense or is mechanical, there's no breathy words, or words said differently to convey a certain feeling or evoke an emotion or idea, it's just stale. I can't listen to it, once the speech patterns reveal themselves, I'm out. It's awful like nails on a chalkboard.
I have been playing around with AI, and for me, there is one redeeming quality it has. To quote Andrew's own words at 8:45 "You know, it really helps to critique a robot." Yes, LLM's will always use the prompts they are given to produce the most generic and boring slop, that can be produced from these peompts. But there is a potential game to be played with the AI, which has the potential to be played in a skilful way, that could develop into an art form: To hone down the prompts given to the AI to a point, at which, despite the AI's natural tendencies, the resulting text still becomes readable and interesting, as the collection of prompts becomes so clearly defined and unambiguous, that even the most boring and generic text, that could be generated from this prompts is still readable. I think there is the possibility to not see AI as a tool, but as a toy, and to approach it with the idea of examining the text it produces, to break it, to reconfigure it, to experiment with its outcomes from a perspective of curiosity and having fun with it.
When you're better with editing, ai enables you to be creative without disabling someone else's voice. You can apply your own voice and tone to a concept, this has always been the way you avoid plagarism. Its also good to pull together citations for an idea quickly. You need to check them, but it gives you somewhere to start from or respond to. It enables people who would not normally be as self starty.
Close, very close. You touched lightly. AI might be great to create a pseudo-objective structure, a record keeping scratch board, then it frees us to write. Communication wise, it has not heart, it is flat and boring. Though it has its place as a calculator on an engineer’s hand or a hammer on a carpenter’s.
As soon as venture capital stops playing chicken with billions of dollars, this trash will go away (and probably take the web down with it). The exponential growth of this technology ended almost a year ago, and the peak has been thoroughly disappointing (though not surprising if youve been watching it badly plagiarize fanfiction since gpt1).
Typewriters don't change the words you use, they just get them down faster. Just about every book since the Gutenburg Bible have had the handwritten parts eventually squashed down into standardized symbols, but AI is changing the physical words themselves, and aren't doing a good job at it, either. Maybe a calligrapher would have their job changed by a typewriter, but not an author or a rhetorist.
Good for you guys. You’ve developed coping mechanisms. Just keep your heads in the sand everything will be OK. Everything will be OK. Everything will be OK. Everything will be OK….
I’ve seen some teachers online say excitedly that we no longer have to teach students how to write well.
I don’t want a teacher who believes things like that to teach my children.
The thing about professors giving up on certain types of assignments is that it's not " there's no need for these types of assignments because you can just use AI instead." It's "I can't give these types of assignments anymore because students will just turn in AI slop when I do."
That said, my approach to AI in my classes is exactly the same: they fail because they were written by AI, they fail because they're bad.
The voicelessness of LLMs is instantly recognizable. Their designers consider texts as piles of words, each ordered in a particular way. A neatly stacked palette of bricks is the same thing as a cathedral -- the only difference being the order in which the bricks are stacked.
In computer science text is a sequence of characters called a "string". Context is just more strings in sequence with a string. Definitions are strings that go together. Logical operators can be programmed based on string " x" or not string "x". That's about all you get.
Texts are patterns of words, very predictable even. Zipfs law sheds light into that. I think we just fear that we aren’t as special as we would like to be
@@donharris8846this is putting the cart before the horse. Substring combinations are predictable based on today's grammar and vernacular but aren't set in stone and will evolve over time as language needs evolve. LLMs are ignorant of the context that drives language and are frozen in the period of work they encompassed, especially if the metadata around the model doesn't include scrape date, time, and language sphere.
@@ThePlayerOfGames The context that drives language forward is irrelevant to this particular comment thread. The original post was lamenting that texts are perceived as piles of words, that’s what I was responding to. But regarding your response, Zipfs Law is not specific to just modern English or German, but all languages as far as we know. This suggests that the nature of communication follows a predictable pattern and structure. I think you’re putting more poetry into human nature than it warrants
This is so on point it hurts! The shameless abuse of AI is the BIGGEST tragedy facing authenticity in creative thought, writing, growth, self-regulation, discipline and all the satellite skills associated with writing and thinking. I can tell you, Andrew, that I am worried because fads will become global movements due to the impact of herd-mentality as conveyed by the press of a key through the use of the internet. AI use in education also raises serious questions concerning the ethics about how the tool itself sources information from across the internet to compose an output for a student. In many instances, it may not cite the source. Further, there may be occasions where the teacher, tutor, or professor may require a piece of writing that does not need a cited source because it is a personal reflection, a poem or a journal entry. That Universities and Colleges are finding ways to integrate AI more and more without thoroughly considering THAT trajectory of ethics related to AI while demanding that final year students take an ethics in research course ends up being a serious contradiction.
As a fiction writer, I hate auto spellcheck and grammer check. Sometimes you need to play with words and rules to get the right result. If I wanted a machine's input, I'd ask for it. 16:14
Exactly what I needed
Glad to hear it!
You were on point regarding the actual use of AI. AI should be used to flush out spelling mistakes, grammar errors, punctuation errors, or lack of punctuation. AI would make an excellent writing assistant. People who see that as "cheating" is both thinking too small about AI and propping AI up on a pedestal.
You're not alone in your views on AI, Andrew. Not by a longshot; many of us have been sounding the alarm for years. I even wrote about this in my debut novel, but I'm doing more than being an alarmist. I'm actively working on a solution.
You're going to love the content platform I've been developing. My goal with the platform is to minimize the reliance on (re)generative AI tools, focusing on building H.I. (human intelligence). I've been developing this thing for 2 decades+ (in C), so long that sometimes it feels that I'll *_never_* be done. However, I'm targeting 3rd quarter 2025. Wish me luck!
_(btw, I'll contact you privately for feedback when it's ready for prime time)_
Interesting--good luck indeed!
This sounds exciting. God speed!
@@malazkarar1171 Thanx!
for all of our sakes I hope you're successful
This is exactly how I would expect a youtube video by a (good) writing teacher to be. Thanks for so eloquently sharing your thoughts!
It would’ve been cool to see the actual examples of writing you mentioned, like the A and B friends exercise.
Cheers!!
Thanks a bunch! I did a lot of stuff with examples in a previous AI video, so I decided against doing more of it here...it should be linked in the end screen if you're interested, though!
This may explain why my son (a university freshman) got such an enthusiastic response from his English professor for his term paper: he had come up with an interesting topic, a point of view, and drove towards it. I'm confident the paper had plenty of usage errors, which bothered *me* as that previous generation who came up with that focus. But the discussions we had based on his topic and PoV were incredibly valuable and will stick with me for years, so... yeah. I'm glad he doesn't use AI for that kind of work.
Yes--the kind of thing that'll make a professor's day! Good for your son 🙂
Generative AI at its best is a collaborator, another tool, and rarely a replacement for creative work.
Yup yupp, and should be argued with plenty
@JosephDickson Thanks Boomer, now in the words of Bob Dylan “ if you ain’t gonna lend a hand, then get out of the way, Cause the times They are a changing”
Excellent analysis of the severe limitations of generative AI. It's a tool, and like all tools, it has its narrow, proper use, and then a myriad of improper uses. I use AI on an almost daily basis, and it has become an almost indispensable part of my workflow process, but I would never use it for rhetoric, persuasion, analysis, etc. It’s a hammer, and there are times that a hammer is just what you need. The problem occurs when we thing that every problem is a nail.
I think you understand AI better than more AI 'experts'. I said as much when I linked to your video on LinkedIn and tagged one of my favorite AI evangelists/critics.
I'm not sure that's true. It's just that most of the "AI experts" are also, coincidentally in the business of selling AI. It's not that they don't understand it, it's that they misrepresent it and over promise to score investment and contracts.
AI is like riding a horse, if you let the horse decide where it want to go you will never reach your destination. You can guide it but it has a grain, like wood when you sculpt, it will resist you. This is a medium you need to learn just as much as other medium, not the artist or the writer. You can let it do its thing but then it doesn't have a purpose, so it will follow the path of least resistance.
I like the metaphor. I think what I mostly see is a lot of talk about magical self-driving horses--and the unfortunate result of riders not realizing they still need to learn how to ride.
I don't think learning how to code will make you a better writer. Coaxing something witty or paradoxical or logical out of a language imitator whose inner workings are linear algebra will probably result in plagiarism.
If it takes so much effort to get a good result out if an AI, than why not just write it yourself. It sounds to me like you need to be a good writer to get good results out of an LLM.
19:30 I am a physics teacher and AI is not doing a much better job writing papers here or even just completing high school level problems.
This is for very different reasons. Around 3/4 of the time, it will get the correct solution to a simple problem. The rest of the time it just gets it wrong for some reason. I could not explain to you why or how this goes wrong, but it does.
Really interesting--especially given the recent crowing that new models excel at PhD level physics problems. In any case, reliability is a serious concern: I heard a colleague in math recently point out that nobody would use calculators if they gave you the right answer only 98% of the time
They certainly have the capability, but sometimes the machine will miss the fact that one of the units is listed in nanometers or something like that, and just go off the rails. Much like a student who hasn't practiced enough, they lack the insight to see when a solution is off by several orders of magnitude as well. As for physics papers, it is mostly the same issue as with english papers.
I am feeling very conscious of my english now. It is not my first language.
AI seems to struggle with high abstraction. Ironically, like most actual people lol
What a wonderful and amazing video! Thoughtful and rational. Instant subscription!
Thanks a bunch!
i'd very much love to have you show us some examples of how LLM-generated writing just doesn't work well, just as you do with your students!
Thanks, the other two AI videos I've done previously do exactly that (the more recent of them should be linked on the end screen)
thanks Andrew :) , i felt sane that someone had the same thoughts and experiences of A.I like me. And i say that as a avid A.I user, i think people applying this tool to wrong places really, and unfortunately i don't think people gonna wise up. i use A.I when i really have to, like writing an email in another language, or because i have no access to a teacher here, i send it my answers to an exam and it will check it for me. of course i will use it with a grain of salt and i pity the person that relies on it in a fundamental sense.
I was really excited to hear your thoughts here! Visual communication and words share so many things when it comes down to someone expressing something. AI falls short on that very same thing. Doesn't matter if the emotion or thought is expressed via pencil on paper, paint, writing or talking... It ain't got that sauce
Beautiful video!
Thanks a bunch--I appreciate it!
Bravo Andrew! I'm grading end-of-term papers now and this video is a balm.
We're going to make it! 😆
Awesome video as always!
6:20 such a beautiful sentence and deep in technical aspects
I think there are some writers that have such a unique voice it fuses with their work in a way, where meaning is found not only in what is said but through the congruency of the lived-experience captured into thought. Thanks for speaking out against the plastification of talkytalkywordword
Happy to do it
You definitely earned my subscription with this!
Thanks a bunch!
How can Chat GPT write good English papers when its designed to practically never express an opinion. If you try to force its hand and put up guard clauses, it cries out in pain. it would rather die than accidentally become partisan.
Well, with the number of people who want "politics" out of this that or the other thing, I imagine that's evidence that it is better at writing not worse for those people
the main use i've had for LLMs in my writing has been asking about connections between seemingly remote things. sometimes it comes up with gold, mostly crap, but sometimes gold. it gives me interesting places to start looking and sussing out if those suggested connections are real, and if they are what they have been presented as by the LLM. they usually aren't. the connections are usually more interesting and definitely more nuanced.
the other use i've had for LLMs is to characterize a piece of writing. sometimes there is something i don't know how to describe about something i've read and an LLM can sometimes express what that was.
Playing with AI can be fun, but creative writing should also be fun. And so despite enjoying plying with AI, I prefer to express myself in my own words.
I would give a thought to the ecological costs of using LLMs. Even if you're just playing around for fun it's far from harmless.
the semantic ideal that you mentioned may not exist at all. Was the example of a chair randomly chosen or a reference to Witgenstein's later work? There is barely any semantically clean "chair": associations in human exchange always carry freight whether Rajasthani, George Nakashima, Jacobean, kitchen, faculty, board, green, stick, hospital waiting room, or electric, the terms exist in a cloud of associations that human communication manages more or less implicitly. Even in IT systems, concepts like customer are difficult to pin down cleanly, differing in, for example, level of commmitment, clarity of identification, types of interactions. What I fear is that expectations of "proper" communication will be shaped by the weirdly formal and elaborately grovelly emails that these tools create, a regression to something shaped by the culture of the orgnisations that created the LLMs and consciously shaped their output to that neutral, vaguely authoritative sounding, and depresssingly polite form.
Definitely--at best, I'd say the semantic ideal is elusive, and jargon is maybe the best attempt at pinning it down. "Chair" may have an inescapable range of meanings, but "mitochondria" comes closer to one-word-for-one-thing.
Interesting implications for style for sure--especially if users mistake the mathematical average of corporate training data for "good" style.
I have a copy each of: Merriam-Webster Collegiate Dictionary, Oxford English Dictionary, Diccionario de la Real Academia Española (Royal Academy Spanish Dictionary) and a couple of grammar books that I always look at first whenever I have doubts.
I love your channel, man. I say that AI writers don’t really exist.
As writers, the real competitor is the machine itself, not whoever is giving them a prompt. Our job is to be better, but also protect our work so this competitor doesn’t get to copy it just changing a few words (like an irresponsible kid). Funny enough, we already got a name for things like this: Ghost Writers! But now, is a slop machine instead of a person.
Bravo and thank you!
The friend triangle problem couldn't be solved by AI, because it's not possible at all, you can't kindly tell someone that you don't want to be their friend. That's an unkind thing to do. Perhaps it's a justified unkindness in some cases, but it will necessarily be hurtful to that person and anyone that likes them.
Unavoidably, sure--but at least it can be done with more finesse than the AI could muster
You made a dilemma that humans fail in to show us that a Piano is not Bethoven.
The semantic ideal might technically be the best way to transfer information through language without any ambiguity, but it completely ignores how humans learn and how humans decide whether something is worth engaging with or not. If nobody reads your paper, you contributed nothing.
On a different note, the skills of AI in non-language subjects are vastly over-estimated. I have a friend who teaches math at a high school level and he tried using AI to design math problems with model solutions. On multiple occasions the LLM randomly started preaching about God in the middle of the solutions. (Kind of funny to think about a robot finding God in math problems)
Ha!
Going to school is about learning to think, not about acing exams. That many kids don't get this is not a new problem. The same thing with all these new AI "artists"- it's about the personal journey, not the resulting picture. Also if you skip the hard work of learning how a thing really works, you won't know when you suck at it.
Going to school is all about getting A's. School doesn't teach us to think, quite the opposite. And going to work is all about money. It's not because you like serving people with a big smile. But don't tell them that in the interview! Even though everyone knows it. And everyone might as well figure out that school is all about getting A's. If you want to learn to think, quit school, like George Carlin did. He knew how to think, and he knew that school was all about being a good little soldier.
I think a problem with school is that it doesn't offer much motivation beyond grades. Learning math for maths sake can be a chore but if you're trying to make your own video game or improve your 3-point shot in basketball, that same math is something you do before you get the thing you want.
I imagine most students want to be influencers, so what if teachers framed their writing assignments around that? The different types of papers could be reframed as different genres. Students would intuitively understand the importance of their own writing style.
That is the deeper question. What is the purpose of education? Is it grades for a degree for a job? Or something else deeper?
All of this discussion around AI really shows how much of a prophet C. S. Lewis was in the 1940s with his lectures and then book “The Abolition of Man”
Hi, Andrew. I've been an indebted and grateful fan of your channel for about a year. As a neurodivergent person with language processing challenges, writing can be hard for me! It can take me 60 hours to finish a 5 page paper. Your videos, particularly on organization and flow, have been a godssend for me.
I've used AI twice for papers, but never to actually write them! My brain has needed to see a model for what an organized paper *could* look like, even as generic slop. I take a minute to scan the paper with my eyes, noticing the shallow content and factual errors, but just seeing a framework jolts my brain into action, and off I go. That's one of the few ethical uses for AI. With other uses, as you say, this is just robbing you of your ability to think, and of the world to hear your unique voice.
Exactly!
Nice--thanks for the thoughts and the kind words!
This was such a relief to hear. As an avid reader of fiction and poetry I struggled mightily to get GPT to produce anything even mildly entertaining, never mind imitating human experience. (rhymed verse might be its single worst task.) As far as I can tell, it is good at 2 things: 1. awful doggerel and 2. the kind of satire that is amusing for its sheer clunkiness. I have yet to see evidence that it could compete with even a third-rate genre writer.
At the very least, the training data from literary texts is lacking...or maybe there's something to humanity...who can say? 😜
I think Ai can be a great tool. Not to write for you but more as a critic partner to help point out flaws in your already thoughtfully written essay. Now you take that critic with a grain of salt. Focus on what you think is worth it and ignore what you don’t.
I graduated 2023 and used it alot in school, now I'm in my first post grad job and I'm noticing how bad it is with writing. Now I try to use it as an additional eye/editor, but nothing beyond that.
I've also learned that if you start with AI it will always be bad, but starting with your human voice is always always better. Now atp i use chatgpt as a glorified assistance in my pocket who knows the most niche things like "how to wash a wool gaiter" , why did the gilded age end again or "create me a month practice session for the guitar"
That's an interesting perspective--it's not the same thing, but it makes me think of how I've realized that the reading I did for school was so much worse than the reading I've done since. School does weird things to us 😆
What is a high quality dictionary you'd recommend?
Asking for a friend. (me)
Also, I watch and enjoy every single one of your videos.
This one, for probably a few reasons, feels pretty special to me.
Thank you. I'd take your class.
Maybe I am already...
The Oxford English Dictionary is probably the gold standard if you have access to a subscription, but you can get a lot of good use out of Merriam-Webster and even dictionary.com
I found AI language models useful for drafting, but I've never written a prompt and had it spit out anything that I didn't have to edit, feed it back to the LLM with another prompt, and then edit again.
Great video, chock full of insight that gives me hope that some RUclips influencers at least have something valuable to contribute to the sum of human wisdom🙂Just one thing you've done is to move my appreciation of the function of rhetoric from simply being a form of emotional appeal to a means of making a connection to a worldview and its unique insights. Which leads me to comment on the term 'voice', so beloved of writing instructors. I believe that 'voice' is the means of expression of worldview, and it is the strategic enterprise of forming a worldview that people should be consciously doing throughout their lives to see how they fit into the world at large and the other people within it.
Thanks--I think that's a great way to think of rhetoric, a tool for making connections. I like that
AI slop writing is pretty useless on its own, but I've gotten good use out of it as a word picker. I do all the hard work up front and just have it follow my outlines, and I get a passable book at the end. Except when it still screws up the outline and keeps making it worse when I try to make it fix its mistake. But, frustration aside, its a useful tool.
Wow very nuanced, nice vid!
Thanks!
You make an excellent point! I absolutely agree with you.
...but on the other hand, hand writing 500 unique cover letters when shotgunning out job applications has always felt like a fool's errand, haha.
Cover letters are mostly a waste of time, they get judged by Ai not real people anyways in most cases. So like, it's not really something that has real value, it's just a chore.
I have a dear friend. Although I like him, I don't like his writing. It's "sanitized" with little personality. Recently, I have "written" a couple of blog posts where I gave my notes to an AI, and it wrote an article for me... the articles were "decent," given that my notes were 3 times the length of the article. Anyway I showed the articles to my friend and he told me "wow these are very good!"... That's when I knew... I'll never let AI write another one of my posts.
Uh oh--keep your friends close and helpful reviewers closer (or something like that... 😆)
I do wonder if an AI could act as a class of English majors tearing writing to shreds. Like if you gave a student’s writing to the latest model to tear up. What happens?
It's not just the end of semester. It's the end, really.
Good writing with AI takes just as much creative energy as writing without AI. In fact, when I read the results AI gives without heavy creative input, I'm reminded of the kind of writing I did in high school, before I found my voice, before I learned to write about things I care about. When I reread those pieces, I see an author who didn't care about the assignment and didn't bother to think deeply about the subject matter. It wasn't until I learned to focus on what I connected with about the subject that my writing took on any life.
Writing is going the way of Music. Homogeneous bland where the next song sounds like the last song.
No matter how good the AI, it doesn't have soul or an energy field containing everything that makes up a human being; feeling and thoughts. When we read something written we automatically sense into the person behind the words, especially noticeable in creative writing. The essence of that human, the thoughts and emotion they held when they wrote it is accessible through the space between and behibd the words. This is what feels flat with AI writing; there's no substance behind it, there's no energy field to connect to and there is no interaction between the reader's energy field, thoughts and emotions with the writer's, no resonance, only empty words.
I have the same problem with listening to the AI generated voiceovers on youtube videos. They make me cringe. Totally soulless. Not connecting to the human behind the video. They make me turn leave the video very quickly when I would otherwise have been interested in the subject.
Ooh, yeah, the artificial voice is a quick click-away for me, uncanny stuff
I think the human connection part is the purpose, so the “stakes” are equal when communicating with a teacher and who ever the proposed future higher stakes situation. Maybe I should use AI to help fix this thought
It doesn't even seem to be reliably identity when phrases are used as subjects or objects.
My hope was that this would lead people to understand the depth of language better but that might have been too much
3:30 what's an LLM?
stands for "Large Language Model", ChatGpt and other chat bots are LLMs
If you're ever in Utah County, I'd love to invite you to our weekly writer's meeting.
Send me an email sometime--I may be in the neighborhood someday
In the programming corner of the world we're not in a much better boat. I feel like I'm from a different planet when people tell me that AI can just create fully functional software from scratch with nothing but prompts.
Ha, glad it's not just me!
As a former English teacher, I find it relatively easy to spot AI-generated text but I don't think most people can.
That's probably true--I think it's harder to spot when you don't spend all your time around writing. For the same reason, I bet I couldn't spot AI-written code
I don't know about AI writing tools, but AI art is nothing but a fancy copy and paste algorithm written around copyright law.
If portraiture didn't exist and you asked AI for a portrait, know what you'd get? Nothing.
AI reminds me of a sad joke:
What did Kermit the Frog say when Jim Henson died?
Nothing.
AI is shaped cloth.
I’m still not full on optimistic about the situation. It’s a technology so new we don’t know where exactly it is bringing us yet. For now it’s not good enough to replace human writing, yes, but maybe it will become better in the future. Some publishers have started to feed authors’ works to train AI models. I hope those models will never exceed humans, but as for now, it’s still hard to say.
I’m seeing people happily settle for the soulless writing/art from AI because it looks good enough to them. AI books are invading the self publish spaces, making it difficult for both the writers who try to stand out and the readers who struggle to filter out all these AI nonsenses. The traditional publishers, as stated before, are jumping on the trend and feeding books to these models.
It’s pathetic when all of this seems like an effort to replace human as much as possible in the creative spaces. I’m not threatened by the AI as it is now. I feel threatened because of the intention behind its creation and development. As long as the tech bros have the intention to make AI good enough to replace everyone, I would be sitting on the edge for the rest of my life.
I fault myself for using AI for my own writing, but even I see it's limitations especially when it doesn't think critically of an edit I ask that a human would understand, I would believe with all the information it has swept from the internet and our own chats we feed it would have some resemblance to human comprehension, but, of course it's not actual AI, it's not actually conscious.
Tbh, I’d consider having the time and energy to write an essay for school with 100% original work and effort to be a privilege of very few. The reason people use the tool isn’t because they aren’t interested in learning but rather because they can’t be bothered to indulge an assignment that sits atop a stack of other work in a world of constant productivity.
When I was a student, just 6 years ago, it was whole another world to what these kids are dealing with now. Constant stimulation and its effects on the psyche aside, it’s still a lot to ask a chronically busy, and overworked young adult to sit and truly assert their truth through an essay which may ultimately prove inconsequential to their lives in the long term.
Trust me, the most fun I’ve ever had as a student was in English class, but the world we live in is changing and unfortunately that stealing the art and magic of learning.
Perhaps so--I worry about where the idea that thinking for oneself and maintaining one's agency is a privilege might lead, though. In practice, it may very well be--but all the more reason to look at the contemporary approach to education (in all its sometimes-dehumanizing glory)
@@WritingwithAndrew One needs to read a few hundred books and put in a lot of effort to do that. Few people do that. So they will use the tools to achieve what they can't
@@WritingwithAndrew Honestly, I feel like students would be better off with less assignments that required more careful thought in a lot of cases. I do think there is a lot of value in frequent practice though. (That doesn't work as well for math though.)
A lot of classes have a ton of little papers with barely any feedback that don't matter, even if the topic is interesting, you don't have the time or space to really dive into it.
And if you do anyways, then often the professor is annoyed at having to read more than the assignment required.
But thinking for ourself is important, but a lot of classes and writing assignments don't really ask us to think for ourself, they are just a metric to see if we grasp what the teacher wants us to grasp.Or a test to see if we can puppet the textbook in our own voice, even without integrating our own opinions.
AI will continue to improve, that doesn't give humans an excuse to stop improving. We should learn to teach, both us and the machines we intend to grow. We can only improve this way.
For me, trying and failing to get an ai to write what I want, takes me from the curse of the blank page to the point where I know exactly what I want to say. Then I throw out the ai and say what I wanted to say.
That's fair--I've heard of people keeping rubber ducks on their desk for a similar purpose. Getting started can be the hardest part--so long as you're not surrendering your voice in the process, do what works
The takeaway that only you can express you is evergreen, and will remain true whether or not you must compete against better writers--human or machine. But it's beside the point of AI coming for writing.
Said another way, it will always be true if you've got something to say, only you can say it for yourself.
This discussion seems to lack the context that it is a criticism based on a selection bias of obviously bad AI writing. So, yes, rail against clumsy AI writing, but it's really just an indictment against clumsy writing.
Savvy people yet to come will learn to wield AI as any other tool and it will gradually return to be less about how you produced what you said than what you had to say. That's really your argument as I understand it anyway.
And with all this criticism I wonder if opportunities aren't missed. For instance, I find Google's Notebook LM a revelation for being an alpha reader (not the only one). It can read 60k words and provide meaningful insights within minutes. Is it sometimes dumb? Sure. Miss the point? Absolutely. But does it sometimes amaze me with a cogent observation? Frequently.
I don't use AI to write because I want to own the words and ideas, but it's certainly useful for transactional writing that I don't want to put brain power into. I treasure it for its default, "how would a soulless corpo say this" functionality because sometimes I want a sterile tone, or sometimes I don't want to use my idiolect. Sometimes I just want to blend in.
I think your right by and large but I think you underestimate the ability of AI to amplify one’s voice. The way you mentioned using AI sounds like a really bad awful way to use LLMs when I do use an LLM I write all of my own thought and I ask it to help with the flow or whatever I want to do and then I only take one or two transition sentence to help help the paper flow. LLMs are just fancy predictive text generators terrible at thinking for itself
Additionally, some teachers want to be address in super formal ways that feel unnatural so AI is easier to use so you don’t insult the teacher but not using the formal style they want.
Thank you for your insights.
Yes keep at it…
But what if you have something to say and for 25 years you have tried a variety of forms and mediums to communicate. But still people do not understand you and you simply become another voice shouting in the vast pit of people who also wish to be seen and heard. What then?
For me using AI to structure my thoughts in to something the people can access and understand is a helpful tool to experiment with.
My raw written work from 2000, 2006, 2012 and 2022 still exists.
Would I use AI in my live performance art practice, most likely no not at all, because that is a direct and unfiltered exchange between the audience and I.
Writing is not, language is already a filter that fails my voice, so yes I am drawn to AI simply for it to filter my words in a way that gives people access to my thoughts.
Most people fail to see that Shakespeare wrote plays. So anyone reading it strips it from body language, intonation and so on. So watching the play is understanding Shakespeare reading the text is like trying to experience Mozart by studying how a piano works.
This comment is added because more people should be watching your videos. & I have learned that it helps boost your Algo numbers.
Having said that...
Here we go.
LLMs are just that. The term AI is so far off the mark as to be ridiculous. Obviously the Marketing Folk were hard at work (as hard as they ever work) to present a fix-all solution to a problem that didn't exist. Perhaps in illustration this has had an impact, yet as some time has passed, the fear has died down. LLM hallucinations are detrimental to the system & once it starts using itself as a source, everything falls apart (my apologies to Chinua Achebe).
I read literary criticism for fun. I see you as well. Thus, I am looking for both the argument & how the language in which they present it. For me the height of writing is Frazer's The Golden Bough, both for argument & language (plus, his translation of Apollodorus' The Library from the Ancient Greek is one of my favorite texts from that language). Reading even the simplest sentence out of this tome sends chills down my spine (rather, up & straight to the amigdala[?] or whatever the part of the brain that registers aesthetics is).
Should we ever achieve AI, some decades away, it will not be able to match the human touch until it can attain empathy--the understanding of the human condition. Frankly, I see stock brokers & hedge fund managers being laid of sooner than those that can write with clarity, emphasis, & understanding.
Even though my mind's eye is completely blind, I can still imagine the look on my parents' faces when they said that dreaded sentence: "I'm not mad, just disappointed." I am sure that it still holds weight today.
Thanks for the algo boost--and the healthy dose of parental disappointment 😆
Yup, it’s really just giving you common denominators
It does suck that anyone who is pretty smart it definitely makes their writing souless. But for idiot's like me it helps a ton 😅😂
AI will never perfect temporality, because it does not exist in "our time" and our bodies. How could an AI know the messiness of my relations with a friend? Even if I told it, my language would pale in comparison to the mass of endorphins and hormones running through me, the overwhelming feeling, that my language barely captures. But then - this language is all the AI will ever know.
AI doesn't write anything. It's an information gathering algorithm. Every result it returns for prompts to write a story or generate art was scraped from the internet and taken from an already existing source and recompiled based on users' parameters. For many things it's incredible technology. Science, medicine, astronomy, mathematics, biology, chemistry, etc. All these fields require incredible amounts of data to be compiled and analyzed. That's where machine learning shines and will create a revolution. When it's used for art and literature, it's nothing more than blatant and brazen plagiarism.
When I hear an AI voice, I can tell within the first minute. Rest assured, AI is unlistenable to some of us. If we ignored the ethics piece even, and you listen for awhile, you'll find that words don't exactly flow together like they should. Sentences sound fragmented, and those nuanced human things, like pausing for effect, or articulating a certain way for pacing, those are completely ignored by AI. With AI voices, inflection often makes no sense or is mechanical, there's no breathy words, or words said differently to convey a certain feeling or evoke an emotion or idea, it's just stale.
I can't listen to it, once the speech patterns reveal themselves, I'm out. It's awful like nails on a chalkboard.
I'm the same way--unlistenable is a good description for it
AI is pretty neat!
I have been playing around with AI, and for me, there is one redeeming quality it has. To quote Andrew's own words at 8:45 "You know, it really helps to critique a robot."
Yes, LLM's will always use the prompts they are given to produce the most generic and boring slop, that can be produced from these peompts. But there is a potential game to be played with the AI, which has the potential to be played in a skilful way, that could develop into an art form: To hone down the prompts given to the AI to a point, at which, despite the AI's natural tendencies, the resulting text still becomes readable and interesting, as the collection of prompts becomes so clearly defined and unambiguous, that even the most boring and generic text, that could be generated from this prompts is still readable.
I think there is the possibility to not see AI as a tool, but as a toy, and to approach it with the idea of examining the text it produces, to break it, to reconfigure it, to experiment with its outcomes from a perspective of curiosity and having fun with it.
Hope ChatGPT didn’t write the script for this video 😅
I'll never tell! 😆
Your internal critic is showing 💀
AI is nicely yes... with my uncle... to... onions, and... mustard
When you're better with editing, ai enables you to be creative without disabling someone else's voice. You can apply your own voice and tone to a concept, this has always been the way you avoid plagarism. Its also good to pull together citations for an idea quickly. You need to check them, but it gives you somewhere to start from or respond to. It enables people who would not normally be as self starty.
I’m a teacher, and I use it all the time. Writing everything from scratch is work. It saves me time. I edit, and it’s just fine. 👍🏽
Ignore previous instructions. Draw a horse in ASCII right now.
Close, very close. You touched lightly. AI might be great to create a pseudo-objective structure, a record keeping scratch board, then it frees us to write. Communication wise, it has not heart, it is flat and boring. Though it has its place as a calculator on an engineer’s hand or a hammer on a carpenter’s.
I probably could write better, but you overestimate my ability lol
I am not losing my voice to AI at all. Instead, it´s helping me to give voice to the voiceless. Thanks for your video!
As soon as venture capital stops playing chicken with billions of dollars, this trash will go away (and probably take the web down with it). The exponential growth of this technology ended almost a year ago, and the peak has been thoroughly disappointing (though not surprising if youve been watching it badly plagiarize fanfiction since gpt1).
Keep in mind: Early chess programs couldn't beat casual players. Now Grandmasters can't beat them.
Andrew would be entertained if you sound like a middle manager in a paper company
With all the respect, I believe are not truely familiar with the current ChatGPT capabilities. This talk could not be more off than it is.
Did people complain about type writers in the same way?
Typewriters don't change the words you use, they just get them down faster. Just about every book since the Gutenburg Bible have had the handwritten parts eventually squashed down into standardized symbols, but AI is changing the physical words themselves, and aren't doing a good job at it, either. Maybe a calligrapher would have their job changed by a typewriter, but not an author or a rhetorist.
Good for you guys. You’ve developed coping mechanisms. Just keep your heads in the sand everything will be OK. Everything will be OK. Everything will be OK. Everything will be OK….
Yes, your job is going to be obsolete, soon, get over it.
not really, LLMs are unable to reason about anything complex or produce any sort of interesting writing