We can save what matters about writing

Random thought that just occurred to me: something I always wanted as a student but which is likely impractical because of the time it takes from the tutor, is to have an exam in the form of a conversation, where you can elaborate and explain and ask follow-up questions when you’re not quite sure, etc. This actually seems like something an LLM chat would be perfect for? Let the student talk to it directly and have it ask about whatever the course is about, and then have the teacher evaluate their ability to reason and explain various concepts.

My feeling is that saving writing from AI is like saving thinking from the pen.<p>If we were to live in a society where building a cohesive thought has to be done completely in one&#x27;s head and communicating it needs to be done via talking, the pen might seem like a crutch that will make people lose those skills.<p>---<p>By the way, I asked GPT-3.5 what it would change in my text above, and it says:<p><pre><code> Change &quot;seem&quot; to &quot;appear&quot; The word &quot;seem&quot; is more informal and less precise than &quot;appear.&quot; By using the word &quot;appear&quot; instead, the sentence sounds more formal and adds more clarity to the writer&#x27;s intent. </code></pre> I&#x27;m not sure I agree. What does everybody think?

I have yet to see a LLM generated piece of writing or art that I like as art or that evokes anything.<p>Art is about communicating human emotion. I feel like I can tell when there’s nothing really on the other end. I don’t think AI is a threat to real capital-A Art. I think it is a threat to any form of content milling since it definitely churns out filler much faster and cheaper and better than humans. It’s also probably a threat to pseudointellectual sophistry since it can do that very well.<p>I wonder if the appearance of genuinely compelling art coming from an autonomous AI might be the most compelling sign of sentience?<p>If so then William Gibson nailed it again. That’s what the final superintelligent AI at the end of Mona Lisa Overdrive (final book in that trilogy) is doing.

What we&#x27;re seeing now is a bit of an arms race between cheaters and teachers. Why don&#x27;t teachers &quot;cheat&quot; a bit themselves?<p>Namely, why don&#x27;t universities extend the statue of limitations on academic misconduct to several years, or even decades? Collect all student assignments in a database and routinely re-run them through the most up-to-date cheater-catching software available. Anyone who gets flagged, even years after graduation, gets their degree revoked if they don&#x27;t make an adequate defence to the university.<p>If this is done, students don&#x27;t just have to fool teachers and their tools at the time they take the course, but all tools developed for years afterwards. Having cheated would become like having a sword of damocles hanging over your head. If you could pour years of tuition and &quot;hard work&quot; into a degree and establish yourself in a profession only to have it all ripped away years later because you cheated, how many would still cheat?<p>With methods like this, the take-home essay could remain a part of education.<p>If this seems too objectionable, there are still labour intensive ways to work around the problem. e.g. Oral exams. You can assign take-home essays and then have the prof or TA&#x27;s interview students about the essay. If they can&#x27;t answer some basic questions about their own essay and some of the sources it uses, that&#x27;s a fail.

it’s just showing your work with more fancy descriptions.<p>if you have a beautiful term paper that emerged from the ether, it’s likely synthetic.<p>if you have notes and outlines and all the trappings of having written a term paper, then at the very least you understand all the components and the process, even if you managed to use a language model for every step.<p>that’s the point, the box is open. but it can be used rather than abused. things change. how we learn is changing.

Edit: This turned into a rant.<p>&quot;Teach the process&quot;<p>Learn to learn<p>etc<p>The approach the author takes is boring and therefore will not succeed. Maybe future will prove me wrong on this - I&#x27;d be fine with it.<p>My take is an axiom that is omnipresent among tech bubble: Focus on what matters. Let&#x27;s ask ourself every time we put more than one thought into something: &quot;What is this really about?&quot; &quot;Why do I actually have to do this?&quot; etc. Even without AI our academic systems are being gamified every day. AI just accelerates (and partly even democratize) this movement. People follow a fassade and sadly the system followed. We need to start asking general questions more again. With AI and art this is even more visible. It is often overlooked that AI can not create something new. And you do not have the right to make a living by rearranging and combinging already existing ideas. But exactly this is the metric we should focus on: the new. If you don&#x27;t want to create something new, then fine, just copy and paste along. But making something new is hard and you&#x27;ll need to learn the basics yourself (tech the process, learn to learn, etc). But to make this approach imposed is even another misguided junction. It has to come from the one thing that is important: the new, the actual drive, the &#x27;doing it because I want it&#x27;

Everytime people came up with a new curriculum or a new way to teach it seems to always backfire. Here&#x27;s a thought, dont do anything. Let education be the farce that it is, save for some people who really believe in it, which are the only benefactors since the beginning anyway.

The shift from print to digital mediums has democratized the act of writing, allowing more voices to be heard and expanding access to information.

We&#x27;re only about a year into the post chat gpt world. We&#x27;ve seen nothing yet.<p>Asking humans to do things manually that a computer can do effortlessly and flawlessly is going to be relatively low value. I remember going to school and having to learn to work out on paper how to multiply and divide numbers. The home work was super tedious and I was bored with it. And right in front of me was the solution: a calculator. This was the early nineteen eighties. I&#x27;m sure I could work out from first principles how that stuff works but I haven&#x27;t multiplied or divided large numbers manually ever since I was allowed to actually use a calculator in high school about 35 years ago. Why would I? It&#x27;s tedious, error prone, slow, and I have multiple tools within reach that can do this for me.<p>The adoption of AI is the same thing. The generation after us will not know a world where AI wasn&#x27;t a thing and will use it without even thinking about it. Effortlessly, skillfully, and effectively. And they&#x27;ll be better off, mostly. Us old timers rambling about the lost art of writing (and a lot of other skills) are basically just like any generation that got confronted with the world changing around them, no longer relevant. Besides, if you&#x27;ve ever had the pleasure of having to grade a bunch of student essays, you&#x27;d realize that most of them are not great writers to begin with. Not having to read a lot of badly written garbage is actually an improvement. And being able to ask an AI to extract the essentials of any human produced drivel is probably a great productivity enhancer as well.<p>In the same way, I don&#x27;t need some second rate lawyer messing up legal work and overcharging me for the privilege when I can have an AI do it right. AIs are passing bar exams now so that doesn&#x27;t seem like an unreasonable expectation. Once it stops hallucinating, we can get rid of the remaining lawyers. I run a startup and I deal with lots of weird and hairy legal issues related to HR, taxes &amp; profit, NDAs, sales contracts, etc. Mostly that still requires getting legal advice from skilled individuals. Our startup is in Germany and I&#x27;m not a native speaker. This shit is super complicated and exactly the kind of stuff where you can get a lot of value out of AIs. We still cross check these things with actual lawyers. But it saves a lot of time, effort, and billable hours if you can get a grip on these things before you talk to them.<p>Artisanally figuring it out all by yourself unassisted by modern tools might be interesting to some as an artistic expression, or for esthetic&#x2F;sentimental reasons, or whatever. But when the end result is mediocre and sub par, it won&#x27;t have a lot of value to others. It&#x27;s like a toddler drawing something. It&#x27;s cute and generally greatly appreciated by parents&#x2F;grand parents&#x2F;etc.. And just because we now have mid journey, doesn&#x27;t mean the toddler&#x27;s drawing has less value for them. But let&#x27;s face it, most toddlers don&#x27;t turn into great artists.