When you lose the ability to write, you also lose some of your ability to think

I read for probably 8-12 hours a day. I write every day. I think every day. I’ve been doing this for decades. GPT has been like a talking dog. It’s an insane achievement, but he’s not the worlds best conversationalist.

But to be 17 again and struggling all night to write a three page paper filled with unoriginal ideas, it would have been a game changer back then.

An example I had last night: how do you split a pdf into separate pages on the command line, and how do you put them back together again (I replaced one of the pages with a scanned signed version).

This is a basic task, a beginner's task in the world of pdf munging, but I didn't know the answer off the top of my head. ChatGPT gave me command lines straight away.

That's what they're good for right now. Beginner / common tasks in areas you're not familiar with.

And a further implication, they're great for one man bands, generalist roles where simple things need to be done in lots of different areas. For specialist deep roles where you have all the tacit knowledge in mind already, not nearly as useful.

Especially in refactoring or working with poor codebases. I do notice with copilot specifically some people seem to get wildly different response times with it, for me it gives suggestions almost instantly so as im typing ive got an ongoing suggestion I can accept at any time as it gets closer to what I intend or ignore if not. A common really useful thing is when I have to set up a series of something (recently in a power plant code I set up parameters for a certain kind of battery) and then I just start writing for each other kind of battery it suggests the whole boilerplate for every case, I can go back and fill in the details and exceptions.

As for chatgpt its more useful at interpreting existing code. I often give it a giant function and say "I want you to split off this part and put it into its own thread, get the output with a queue" or something like that. Again, output can often need a little bit of massaging, but I find it reduces cognitive load involved in doing such a task. The trick is to be naturally skeptical of the output but appreciate how much of your life is spent writing nonsense.

I've also found that learning to put what I want exactly into words can be a particular skill that has helped my communication. Making sure you know the name and function of every tool you use. I think if I had said that to myself before I started using these things id have been incredulous, "of course I know the name of things" but no, its a different thing to be talking about them often, especially when you have an idiot savant that takes everything you say too literally. You have to learn to be consise and precise

I can see it guide me in the right direction if I have a more generic question but even then, if I am not specific enough I fear I might go down a rabbit hole that ends no where.

It may be better than google in finding something but when I manage to find something on google the information is usually correct and not totally wrong. Especially technical problem that someone wrote somewhere on a forum regarding a specific device.

Finding the things I am looking for on google allows me to judge the correctness of the data. When I find answers in what is clearly blog spam I know that the info is probably wrong. Which ChatGPT I have no idea how it arrived at the conclusion it gives me.

A good example is if I want to know the max current rating for a chip. If I search google and find the manufacturers datasheet, I know the information is correct. If however the same information is on stack overflow or some forum I may not take it as a fact but more as a guidance.

ChatGPT would be 100 times more useful is it cited a source. Like "The max rating of chip ... is ... according to the manufactures datasheet dated ... which you can find here ...". I guess Bing and Google are trying to do something like that.

The whole thing reminds me of the blockchain hype train. Still using and loving databases here for the foreseeable—still writing things by hand for the foreseeable, and loving every moment.

With prose it's all your manual work again to check. If your text is mundane it might be trivial but for anything a pg would write I guess an AI is pretty useless in its current state.

1. Use it when you know how to do something but don't want to.

2. Use it when you don't know something, and don't know how to find the answer.

3. Never trust. Verify.

Clearly, this means the examples that impress others will often not impress you.

Most people are terrible writers, and never read anything at all. So they have no way to judge if the wordy informationless crap gpt wrote is good or not. It's long, thus is good.

It is not revolutionary for areas you specialize in, but if I try using it for something like setting up a web scraper with python (something I understand conceptually but have never done) it feels pretty amazing.

My one concern is that if I use it for JS I can immediately spot issues with it, but I am too ignorant of python to spot subtle problems.

I came to the sad conclusion that perhaps what I write and what I do is not as repetitive or as generic as others. Over the years when I have had to do more repetition, writing once well and ctrl-c / ctrl-v.

The differentiator is, I guess, expectations: I fully expect my own code to strike a certain balance of readability / optimization / correctness. In this context all but the most trivial code suggestions, sloppy as they usually come, instantly fall out of place.

On the other hand, there is a group of people (e.g. those spamming GitHub pull requests...) who expect the output to be kind of visually similar to some code. When they realize that the result also sometimes works and gives sane-looking results, they are absolutely blown away.

Nonetheless it's better than Google at many things. I find it really good at giving paragraph-long introductions to things, instead of pages-long blogspam. It's also good at suggesting improvements when writing in a second language.

But I don't see what those AI startups see when they want to replace all sorts of skilled workers with this bullshit generator.

The trick is to use it for what it's good at.

For example, I had this vague, fuzzy memory of a talk I saw online about a decade ago. It was about a database modelling paradigm. All I remembered was that the country it was developed in was in northern Europe, the methodology was based on splitting out columns as individual objects, and that it had funny little icons such as diamonds on the edges connecting things together in a graph.

Good luck finding that on Google. I had tried and failed. It was bugging me, because I needed something like it for my work, but for the life of me I couldn't dig up the reference.

Chat GPT 4 found it just three prompts: https://www.anchormodeling.com/

That's insane.

It only needs to save 10 minutes of my time a month for it to be worth the $20 - I would estimate it saves hours.

My go-to example is if conditions. When I start with the first case, it almost always figures out the rest of the cases correctly.

Couple hours work done in a few minutes.

I agree that there is currently way too much hype. I mean GPT4 is certainly cool and probably going to be useful in some limited use cases, but it's not the new electricity and it's not going to take over the world.

But in the end, it's just a tool. If it isn't working for you, it isn't working for you. For me, I prefer thinking in what structure to me as bigger building blocks.

So I prefer vim bindings because I used up teenage time learning it. And now I don't think in terms of characters but in terms of text-blocks.

I prefer writing Java in IntelliJ and Rust in Clion because I don't think in char terms since I can use the built-in refactor tools. E.g. I write functions inline and then extract them rather than writing them as functions and calling.

Likewise when I'm at the shell I think of pipelines as less than pieces. Being an experienced shell user I can usually write a parallel pipeline with conditionals and loops correct first time verbally with no computer usually.

But my copilot flow is improved even better. I talk to the computer in components:

# get all private ip addresses in Alibaba region cn-zhangjiakou ips=...prompted text... # ssh to each and check last long line ...Prompted text...

It's faster than me and this is great for r-i-search with fzf.

Overall, quite happy. But just a tool.

I’d love to see which of the following might have any correlations with reported usefulness of chatgpt/copilot:

1. How articulate a person is.

2. How much of an expert they are in what they are trying to do with the tool.

3. How successful they are at giving instructions to a human to perform the same task.

4. How much experience they have managing/coaching junior devs/interne/newbies.

5. How much experience someone has at decomposing a problem into smaller parts and identifying the simple parts and complicated parts

There are huge consequences to either answer to the question: “Is using an AI tool effectively a coachable skill?”. I’m sure someone has already looked into this or is looking into this - if it turns out to be a coachable skill, and we can identify what those are, there’s a lot of money to be made in bringing a coursework to market.

Give it a Django model and ask it to create a data visualisation. It'l perform complex ORM queries with filters, annotations and aggregations, then load it into a pandas dataframe, then create more columns in the dataframe, does more aggregations, then spits out matplotlib code to draw the chart. The time typing the prompt is time I would have needed even in the pre-GPT-era, since I would have to think about what I want to achieve anyways, now I'm typing it into ChatGPT, which makes the thinking process more structured as well!

I have built more tools in the last three weeks (for work and personal) than I have in the last three years combined. All thanks to OpenAI's brain constantly pushing me into the right direction.

"Hype" can also mean marketing.

Copilot wasn’t useful for me because I want a back and forth conversation to straighten out concepts. With Copilot you don’t get that and it just sprays out little messes of code.

I use GPT4 like a rubber duck in the sense of “rubber duck debugging” and it can be surprisingly supple and creative in its solutions… many bad ones, but again, push back on those and you get to good ones.

If you asked chatgpt to generate a document, its generates one that reads well but has terrible content. Eg. Super broad or just plain contradictory. But given some not so well written piece of writing, it can clean that up fairly well.

The genius of it is that you could give it the notes of a meeting and get it to rewrite them in a different form. Nothing else can handle language like that.

I only use the free 3.5+ ChatGPT and it's written some nice mapping code and some fun poems for me. Poem subjects ranged from rheology to texas hold'em.

I use chatgpt everyday and am getting better and better at asking questions to get the answers I want.

I used copilot when it was free, and it was best at boilerplate, and sometimes not bad at auto completing the next few lines, I didn't think it was worth paying for at the time. I use Codeium now, that works in a similar way, probably not quite as good as copilot in that it doesn't understand as many languages, but still does a reasonable job with boilerplate type code.

GPT-4 Passed the Bar Exam, so I think there's something to it https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4389233

Lawyer and law professor here: No, no, NO!

Wall-of-words text is a sign of a L.O.A.D. — a Lazy Or Arrogant Drafter.

I teach contract drafting to 3Ls and stress The Two Great Rules of Readability:

1. SSSSP: Short, Simple, Single-Subject Paragraphs. https://toedtclassnotes.site44.com/Notes-on-Contract-Draftin...

2. BLUF: Bottom Line Up Front. https://toedtclassnotes.site44.com/Notes-on-Contract-Draftin...

Also a quotation link would be appreciated.

LAWYERS - needs walls of text

Jeff Bezos is known for his long-term thinking and his annual letters to shareholders are famous for their extensive detail and length.

Mark Zuckerberg is known for his long-term vision and often communicates his ideas in lengthy blog posts and public statements.

ACCOUNTANTS - needs numbers, spreadsheets

Sundar Pichai has an engineering background and tends to rely on data and analysis in decision-making.

Satya Nadella has an engineering background and has emphasized the importance of data-driven decision-making at Microsoft.

Warren Buffett is famous for his value investing approach, which emphasizes analyzing financial statements and data to identify undervalued companies.

ARCHITECTS - needs diagrams, visuals

Tim Cook is known for his attention to design and user experience, and Apple's keynotes and product launches often feature detailed visuals and product demonstrations.

Elon Musk frequently communicates his ideas through visually compelling presentations, and is known for his ambitious plans for space exploration and sustainable energy.

ACTORS - needs a conversation, verbal

Mark Cuban is known for his outgoing and talkative personality. He often engages in conversations with fans and customers on social media and in public appearances, and is very comfortable in front of the camera.

Richard Branson is known for his flamboyant and charismatic personality. He often appears in public events and media interviews, and is known for his storytelling and ability to captivate audiences with his words.

It turns out that was a greater benefit that to some extent makes up for losing the exceptionally well trained memory of scholars from a verbal tradition. Not that that last part isn't a loss though, and his warning against false wisdom is more relevant than ever today when it feels like half the people you talk to online are just googling up wikipedia articles to base their world view upon on the fly.

--edit-- sometimes me not grammar good

However, writing for thinking and writing for storage is two different things, and it activates different parts of the brain.

I'm an avid pen and paper user, and using a real pen on a paper allows me to think much more deeply. It regulates thinking speed, so things doesn't escape or my focus doesn't get derailed. Also it changes how brain works so it can think better.

Also, blogging and writing documentation at work made me a much better thinker, because converting these abstract concepts to meaningful sentences with a context and easy to understand structure affects also how clearly you communicate in other parts of your life.

Offloading this really robs you of the joys and power of meaningful communication on many mediums.

I'd just like to point out that Plato very rarely wrote in his own voice so it's very hard to say if it's his views or not that are being expressed.

In this case however, this is almost certainly an expression of Socrates' views, not Plato's. Not only because it's in the voice of Socrates but also by what's transparent in their actions: Socrates didn't leave anything in writing and Plato left us arguably the most important written corpus of classic Greek philosophy.

Maybe he felt ambivalent about it, but he certainly thought there was a value in the writing.

We measured intelligence by the things you knew. We still have shows with questions based on this concept. This concept no longer makes sense in the modern world. But it used to.

Writing is a crutch, Internet even more so, and AI even more so. Eventually you can build the entire thinker out of those crutches and you need no humans anymore.

"What is the cause of people forgetting texts they had learned before?" I can't remember the precise answer but that was one such line in the texts.

Today if you say that you sound kinda crazy.`

but like socrates there's still a few orders of wandering mendicant homeless-like orders out there (the theravada forest tradition being an example)

I tried asking ChatGTP what it thought and it came back with

>... it's also important to note that technology is a tool, and its impact largely depends on how it's used. For instance, AI writing tools ... can be used to augment human creativity and productivity, rather than replace it. These tools can help writers brainstorm ideas, overcome writer's block, write more quickly, and even learn to write better by providing examples and suggestions.

>Moreover, the use of AI tools might also stimulate new forms of thinking. Just as calculators didn't eliminate the need for mathematical understanding but rather allowed for more complex problems to be tackled, AI writing tools could help individuals refine and expand their thinking, enabling them to tackle more complex writing tasks or express ideas more effectively.

Which is kind of interesting - I didn't think of writers block on my own for example. Also lazy thinks at the moment probably tend to just go along with what they see in the media. AI aided thinking could improve on that.

* learning to ask good questions

* learning to fact check

* learning to edit the writing of others

Then they have truly learned the Socratic method. Which is a much more valuable tool than writing from scratch solipsistically: being able to actively engage with the writing of others.

If you replace intense IRL dialogue and discussion, with writing, this might be the case.

On the other hand, if you replace just letting your thoughts fade in the void, with writing them down as coherent though out ideas in writing, I would argue that the writing has actually both triggered retention from memory (thus exercising the memory), and helped you critically assess and make your own thinking clearer.

Thus, I think writing is an extremely helpful tool for processing your thoughts for most knowledge workers who are not already involved in constant dialogue with others (which is most of us I guess).

You know, Phaedrus, that is the strange thing about writing, which makes it truly correspond to painting. The painter’s products stand before us as though they were alive. But if you question them, they maintain a most majestic silence. It is the same with written words. They seem to talk to you as though they were intelligent, but if you ask them anything about what they say from a desire to be instructed they go on telling just the same thing forever.

Because it is now not the case, and AI trained on a corpus of many books can in fact say something new.

That and Meno's paradox!

Consider also Derrida's thoughts on these matters: https://en.wikipedia.org/wiki/Pharmakon_(philosophy)

Especially for this quote, I like attributing it to Socrates as Plato did. We obviously don't know the details and nuances. However, I feel it captures the (perhaps inexact) tension between Socrates' old, mostly oral tradition of philosophy and Plato's newer tradition of written philosophy.

It's ancient Marshall McLuhan. Oral philosophy and written philosophy create different philosophies, not just a different medium for the same philosophy.

It embeds what is remembered in a current context.

Did Plato take this into account?

That is correct. I mean, some of that is exactly why I write at least 5000 words a day of journal notes while working. Because I want to offload everything from memory to a written form that is indexed.

> and you offer your pupils the appearance of wisdom, not true wisdom, for they will read many things without instruction and will therefore seem to know many things, when they are for the most part ignorant...

This part is where we veer into prediction of how it will alter people and it is both right and wrong. I think the education system has shown us it's possible to have a great memory and to remember many facts and still be ignorant. At the same time, there's so much material that can be consumed and people have surface knowledge of many things and once you dig into a conversation with them on it, the ignorance pops up and it turns out that they don't know much about the topic. I know this. This has been me on many occasions and is still me on some occasion.

>... and hard to get along with, since they are not wise, but only appear wise.

Again, there's some truth to this. I think about the times where I've jumped into a conversation about politics with some half baked knowledge that I read somewhere but failed to understand or recall correctly and I think I must have a looked a fool at those times.

But then again, I've course corrected here and there and I've spent time writing and sharing thoughts with others where I've used previous writings to pull up facts and my own synthesis, and I feel like I've been a better participant at times like that. In that way, writing has helped me steer away from ignorance.

I can't comment further though on the quote because I've never read the rest of the context so I may be missing out on some stuff.

I think Paul is on to something here though. I can't count how many thoughts I've had that I thought were clever but when I started to write the thoughts down and organize it on my own, I discovered for myself that either I didn't have enough evidence to prove myself entirely or that I was just flat out wrong. Writing is a form of thinking in my opinion. And if we don't write, we do lose that side of thinking too. I do appreciate though that Paul doesn't go further into predicting the consequences as Plato's warning does here. But I'll be willing to bet that Paul is right that there will be some negative consequences for sure.

I'm not going to deny it's a change, but it's too early to decide if it's a bad change. Every generation has its own big shift in the past 100-200 years, and every generation complains about the change.

In my experience writing is a very good way to do that, but if verbally works better for some then that's fine too.

In that sense I do agree with Graham that "I have some thoughts about X and I'll let ChatGPT take care of the details" is perhaps not entirely ideal.

All of this also applies to code by the way, in more or less the same way.

Imagine the value a professional teacher would have provided for them...

One can tell from the opening of your comment. Composition, cadence, vocabulary, repetition. It ties together.

I prompted an LLM to produce an enthusiastic expression around writing, but it came up with drivel like

> Writing fills my heart with pure delight, igniting my soul and inspiring my mind with boundless creativity.

I found myself tempted to instruct it on how to write well but that sort of defeats the point.

And maybe I suck at prompting but the appropriate combination of profanity, brevity, yet seriousness was hard for me to get it to do.

I don't think marketing copy deserves to be grouped with 'boilerplate' and 'canned responses'.

It takes a lot of brain power to create marketing copy that isn't boilerplate / canned / generic.

Good marketing copy persuades people to take a specific action, and figuring out what this takes just as much critical thinking, iteration, and getting your reps in as any other compelling content.

Which is why most marketing copy feels bland / generic / boilerplate / canned.

It's not easy.

As a lazy writer who never skips legs day, I appreciate this comment. It inspires me to write more than snarky snippets online.

It only takes one or two manic episodes to realize that they're not always super correlated.

> If men learn this, it will implant forgetfulness in their souls; they will cease to exercise memory because they rely on that which is written, calling things to remembrance no longer from within themselves, but by means of external marks. What you have discovered is a recipe not for memory, but for reminder. And it is no true wisdom that you offer your disciples, but only its semblance, for by telling them of many things without teaching them you will make them seem to know much, while for the most part they know nothing, and as men filled, not with wisdom, but with the conceit of wisdom, they will be a burden to their fellows.

- Plato

Although this is different, because it is still written, just not by oneself, which probably makes the effect of relying on it and forgetting even worse.

----

Righting. verb.

1. writing to change the way the world is oriented

> I resolve to continue righting until the wrongs go away.

2. writing to change the way oneself is oriented

> I resolve to continue righting until I'm not wrong anymore.

----

Ambiguity in usage being more or less the point.

"Tools are the subtlest of traps. We become reliant upon them and in their absence we are vulnerable, weak, defenseless..."

The problem is geography. France, Britain, Germany and such were never part of the ancient world. Rome had colonies there. But, these were (relatively) short lived outposts of civilisation. Civilisation was never a dominant culture in non-Mediterranean "Europe."

In the East, civilisation existed for thousands of years before Rome. They always had roads, kingdoms, bureaucracy, restaurants, banks and such. Even in dark ages, after collapse... these were collapsed civilisations. They didn't revert to tribal, oral societies.

The european Renaissance (excl Italy) was never a Renaissance. It's the actual birth of civilisation in europe. It's the first time writing became a major thing.

My other comment almost perfectly fits here too, so I won't repeat it in full[1], but only by doing the work yourself will you spot flaws and errors in your arguments; ChatGPT can't do that for you.

[1]: https://news.ycombinator.com/item?id=35913419

The other thing that I have noticed is that a great deal of the art from the time is wonky and out of proportion. I was recently at the Musée Cluny in Paris and it was really striking how much of the art in there is pretty badly made, as if by very inept creators. La Dame à la licorne stands out in part because it's not that badly made (though mainly because it's an epic multi-piece tapestry that you'll never see the like of anywhere else). I also remember a statue of a saint whose creator was named (I don't remember the name but it was something like "Le maitre de St. Joseph" or perhaps he had the name of a city) and that was obviously head and shoulders above everything else around it in terms of craftsmanship, which had an almost modern quality.

Then I went to the Louvre and I noticed again how the art before and after the middle ages very clearly lacks this shoddy and haphazard quality of European middle ages art. The Greeks in particular practically worshipped proportion, and the Italian masters would not be caught dead painting an unbalanced figure, by the looks of it. But also Egyptians and Assyro-Babylonians, and all the earlier cultures seemed to take well-crafted art pretty seriously (the bad artists were probably flayed alive early on ... brrr).

So something definitely happened in the middle ages that did not do good to art and culture. Whether it had any impact in everyday life, the ability of people to produce food, build houses, etc, I don't know. There's certainly plenty of architecture that's survived from that time, so clearly not everything deteriorated.

Btw, I think it was the same in the Eastern middle ages too. You can see slap-dash ecclesiastical icons in Byzantine churches of the time. And when you visit Ravenna, and see the mosaics in St Vitale (which are from the very early middle ages), it is clear that something was lost in the ensuing years.

Perhaps we're about to sink into a new "dark age" (regardless of how "dark" was the original "dark ages"). Maybe we'll get lucky this time too, and the light will not be forever extinguished.

In my own life I was terrible at arithmetic, and I grew up at a time when memorizing the multiplication tables was a key life skill. I couldn’t to save my life. I strongly believed I was bad at math and struggled through on sheer willpower and mediocre grades, until I went back to college after I couldn’t make it through calc 1 and dropped out a decade earlier. This time they allowed calculators, graphing at that. I got top marks in every math class from then on and graduated summa cum laude in a top engineering computer science school.

How many people have been marginalized because they couldn’t express their thoughts as clearly and compellingly as a better writer? Maybe a leveling of the fields away from mechanical abilities and towards completeness of thought isn’t a bad thing? Will people lean on it and lose what has been a core skill? Some will no doubt. But will it hurt them? Only as much as it hurts to use a calculator to do your menial math.

I think you can say this about people who haven’t learned to write well. ChatGPT seems like a great thing because they can get it to excrete a content-like substance that reads well, and call it writing.

Prompting ChatGPT to produce a text lets you move faster when you have an idea and want to cut to the finish.

The problem is that part of writing is starting with an idea, thinking about how to express it, and then realizing that your initial idea might be flawed or need work.

Sure - ChatGPT seems fine for spitting out web copy or something where the role of the writer isn’t to examine an idea, but just to go from start to done as quickly as possible.

But I think he’s talking about the types of writing that involve wrestling with an idea and, for example, trying to persuade others. ChatGPT will happily spit out copy in that form. But it won’t trip over a paragraph and say “hey, shit, now that I write this out, I think it’s wrong or not well-expressed.”

People bemoan co-workers who coast by on people skills and bullshit over people who have expertise and do the work. ChatGPT is likely to be similar - helping people who can’t write (express or work with ideas effectively) sound authoritative and generate more content that they don’t really even understand.

Control is an illusion.

But that also means that illusions can be used to gain at least some (sense of] control. See that mismatched bracket? I did that, despite the Universe screaming with a thousand voices that I shouldn't.

This text was written by a humane life form. #NoAI

Even formulaic messages and emails can be useful, because when what I want to write doesn't plug in neatly into the formula that sometimes means something.

----

As a very concrete, though small, example: I have found that almost always when I use the word "empower" in any of its shapes I'm making a mistake (either taking a lazy shortcut in writing, or an actual cognitive mistake), yet I only notice it as I'm putting the symbols for it down.

Can attest. I had to write a bunch of definitions and do some preliminary research though nothing very technical. I asked chatGPT, it listed a bunch of stuff, I looked around the net, found some more hints, asked again; then I had the model to collect everything for me into a neat little markdown doc, and shared that.

Saved me a couple of hours of searching and writing useless / boring stuff.

The mere act of writing, regardless of the subject, seems to exercise the rational part of the brain and calm us down a bit. (The effectiveness of cognitive behavioral therapy seems to support this idea, though the existence of Twitter does not...) I think we have to at least entertain the possibility that a post-reading/writing world could be a much less rational place.

[1] https://www.gettingsmart.com/2016/03/26/exercising-student-b...

It's mediocre, generally lacks nuance, and is pretty boring for the most part. On the other hand, it's grammatical and, in my experience, mostly not actively wrong. So as paulg says, that's at least no worse than a lot of stuff I skim over from people cranking out thousands of words a day for peanuts.

I think the pg comment presupposes that most people currently know how to write, and I just don’t think that’s the case, at all. At least not to the extent that doing less writing (or using an AI) will dramatically reshape their modes of thought (much less the trajectory of human thought as a whole).

Also, I’m not saying that written communication is the be-all, end-all of human intellect - I think I’m a fairly good writer (used to be better, though) but a fairly poor verbal communicator. I know lots of people who are great verbal communicators but not-great not-terrible writers. Some people are great at both, and some people aren’t good at either. But it’s all part of the mix of original human thought.

I guess when we get to the stage that a 21st century Orwell is using AI to write his books, then I’ll worry that we’re stunting human potential, but until then it just doesn’t seem to me like much original thought it getting displaced here. Yet, anyway.

It might be better, because the long-winding and formulaic text from the models is something also only models can read well :)

A implies B is equivalent to Not B implies Not A.

The issue is that (Not gain) and (Lose) aren't exact opposites.

The insight is interesting though.

When you give the LLM an input and you get an output, has it gained the ability to understand or think? No.

But when the going gets really tough, those people may try to band together and secede[^1].

One thing I have to add though, as somebody who loves to write, is that I couldn't do it without our modern digital tools. Because I'm not a native speaker, I tend to bungle prepositions and semantics, and it's only with great effort and many Google searches and ChatGPTs and grammar checkers that I dare to afflict[^2] the world with my words.

[^1]: I found this idea in "Homo Deus," blame Dr. Harari!

[^2]: "Afflict": To the point, I used thesaurus.com here to find a suitable synonym of "torment".

Not to derail, but: fat in food isn't what leads to fat people. As far as diet, the culprit is too many calories, and particularly foods which lead people to overconsume calories for various reasons.

And it has a very serious consequence: "when you lose the ability to write, you lose the ability to create new content on which future LLMs are going to be trained".

One guy has been exposed for producing one computer programming book per week, for six months or a year straight, all with 5 stars reviews.

How many people like him are gaming, for their own benefits, the system?

It's all too obvious on GitHub and StackOverflow already: people gaming the system with automatically generated code that just, to reuse you words, "rely on shuffling existing versions".

How can this LLM-generated content possibly be used to created a better LLM?

Basically: how do we train future LLMs using content produced by those among the humans whom didn't lose their ability to think?

It’s the allegory of the cave all over again, mathematics has become a shadow of itself in many people’s minds. Instead of a tool in which to express ideas, it just became stuffy definitions and arcane symbols which produce hard yes/no answers. Why? Because when those ideas are expressed by writing down the mathematics in textbooks, they are expressed in arcane symbols and stuffy definitions which have exercises that have hard yes/no answers.

Everything from email to blogs to podcasts to even discussion forums turned into content mill venues of spam. Any new type of digital medium should absolutely be aware of human incentives, because your intentions don’t matter. People have very limited intuition about game theory type of things, even the smart ones who want to build cool shit.

You and I wouldn't notice if that entire industry was replaced by ChatGPT because nobody reads that text, it doesn't exist to be read, it exists to tickle some ranking model deep in the bowels of Google. But you can be sure that the people selling coherent text for $0.01/word will notice when ChatGPT can generate the same thing for $0.000001/word.

- being able to organize and compose your thoughts

- being able to synthesize disparate ideas into a cohesive whole

- tuning word choice, diction, and flow for a particular audience

- choosing the narrative frame in which you present your idea (this one is huge)

More concerning than long-term decline is that the next generation never develop these at all.

It also means that now we seriously lack people capable of working close to the bare metal - and the same will happen to many other skills and niche knowledge that AI will be able to replace (e.g. why school statics engineers, if AI can solve the math for architects directly).

That will IMHO result in some bad things as well:

- There will be no one left to oversee and improve the work done by AI,

- It will further slow down the overall progress because the most of people will lack deeper insights into how things work and AI will probably still be of the limited use for making ground-breaking original discoveries,

- And finally, though as individuals we will probably keep our IQs, just entertaining them with something else, as whole our civilization will loose some knowledge and skills - making us all dumber in a way.

That’s a skill I’ve developed over the years, and I’ll still need to use it, but it’s often a major time sink. Skipping it lets me focus far more on learning and applying rather than grasping in the dark for a while, digging, sorting through sources, verifying and validating, etc.

GPT will even surface studies pertaining to very specific topics by linking regular language to more sophisticated, field-specific jargon. I wouldn’t use it to learn what’s in the papers, but it’s extremely useful for simply finding what I need to know.

Agreed, but so much depends on where that focus goes instead. Whatever you do instead of writing is going to be what you improve at, while those other skills atrophy. That thing you spend focus on could be thinking deeply about abstract concepts, or composing sonnets, but it could also be mindlessly scrolling TikTok. Whether you ultimately buy Graham's thesis here depends on what you think the most likely outcome is.

My n=1 experiment with driving a manual for 40 years in urban environments coast to coast is that I can't focus well without it and my experience doing half of that data set on a motorcycle just enforces that conclusion.

IMO, this opinion deserves some thought even though it is not as bad as it sounds

But they're not expressing other very specific intents like going out of their way to assure the team that the new opportunity was unmissable or pointedly thanking some people but not others or being deliberately terse because frankly the whole experience was a bit shit. (and might if they're not careful be expressing particular intents they specifically don't wish to convey)

And ChatGPT itself generally gives the most verbose answer possible, because people training it reward verbose and somewhat ambiguous responses more than clear but not quite right answers, so I suspect its impact on prose norms will be the exact opposite...

I can see that

That's simply the world moving forward. You might assume that a house is a smoky place before folks invented the chimney. Horses were important before mechanical transportation got off the ground. Belief in humors has dwindled since germ theory proved to be successful. Reading and non-religious texts being easily available introduced ideas folks might not have had on their own. The camera caused folks to reconsider what art is and allowed for impressionism, abstract art, and more.

Tech and advancement changes the way we think and what we think, and it shouldn't be surprising.

It might just be me, but I find it weird that one would consider the act of actually playing music not to be precisely the fun bit.

Of course writing is important. Writing free-hand on paper is beautiful. But so are the ideas, and the organization of the writing, and the style, and the plot and the characters.

Of course playing instruments is important, but I'm reminded of the Rick Rubin interview where he talks about how he's not a musician, and he doesn't know how to do any real technical music production.

That isn't born out of passion.

The strange thing about being human is that we have an inordinate capacity to normalize and then universalize just about anything. On one had this plasticity allows us to deal with almost anything we are apt to encounter. On the other hand, it's an enormous blind spot to what could be. We're much more likely to add to our initial mental model than conceive of an entirely new perspective and it's these entirely new perspectives that I value the most.

Ironic.

I think you can dunk on someone's writing without ableism. (Observing that a blind and deaf person isn't likely to enjoy a Broadway show isn't necessarily ableist, but gratuitously bringing them into a discussion that doesn't involve blindness, deafness, or Broadway shows is.)

http://lemonodor.com/archives/001091.html

A tool is an extension of man, and it is not additive as you suggest, even for “talented writers”. A tool interacts with man’s mind and body, and society’s fabric, “adjusting the sense ratios” of man, to use McLuhan’s phrase.

“Just a tool” is lazy thinking that is most effectively used to stifle debate about technology. This has happened with the typewriter, the gun, the computer. For a more sophisticated engagement with technology, see Neil Postman, Ivan Illic, Marshall McLuhan, even Frank Herbert.

Which is fine when some experienced person uses the generator to do something that they already know how to do (boilerplate code generation) or is tangential to their main activity (report generation).

In the long run though, what happens when we reach a generation that is taught to use the LLMs for their job from the start? Will they be able to handle anything unexpected?

This is exactly the thought that puts me in your category of "people [who] find writing very difficult"