https://futureofcoding.org/ logo
Join Slack
Powered by
Venkatesh Rao <in defense of slop>.
# of-ai
k
Venkatesh Rao in defense of slop.
🔥 2
i
Just because it was invented and intended as a witch-hunting tool doesn’t mean you can’t use it as a remedial literacy aid.
Could do without this punching-down at people who haven't developed the "literacies necessary for reading AI-assisted writing", especially given that despite describing so many other aspects in tedious detail, these "literacies" aren't explained at all.
So I don’t blame the scared villagers for clinging to their ineffectual witch-hunting practices.
Good to know this person doesn't care about any of the negative externalities of AI. They just found an analogy they like for cancel culture and are clapping back with it. Gross.
Setting aside the endemic culture wars over AI use
red flag
alright. interesting read. not sure what to take away from it, other than: it was hard for me to notice the bits that were authored by ai. despite saying that the particular stylistic quirks of ai were not rewritten away, i noticed none of them.
i also didn't feel the least bit enticed to go read any more of this substack, so i guess that's something. the writing didn't spark and particular emotions, either. quite dry and lifeless.
👆 2
not sure where to place the blame for that — perhaps that's deliberate on the part of the author.
k
My wife says the same about Venkat's writing 😄 This particular piece was paradoxically completely manual, apparently.
i
wild
k
Example of his recent AI-generated work. It's missing the most obvious verbal tics of AI, but the sentence structure seemed very monotonous to even my untrained eye. Was quite exhausting to read though it started strong. The image is quite awesome IMO in spite of being clearly AI-generated.
d
As someone who can describe themselves as anti-humanist probably, I was hoping for a better defence that "edgy".
k
I have never been able to read any of Venkatesh Rao's writing to the end, even long before LLMs even existed. I tried a few paragraphs of this one: same impression. I have no idea what turns me off, but there is something.
💯 1
😆 2
c
One thing these LLMS do: they blur our line of: WHAT CAN WE ACTUALLY PROCESS The seduction of consuming ever more and more, just one more summary ...
As we mirror ourselves in technology its probably a thing that when VEnkats chats a lot that his style becomes and consciously or unconsciously emulates the LLM style.
The strangeness of our times!
w
Overwhelmed... by... look... all this kind of stuff roused me from my academic armchair this year, and it was a comfy chair. I've spent my time trying to make sense of generative AI texts. Look, people acquire language from running around as toddlers. LLMs acquire it from statistical regularities of stuff people post online. That LLM generated text (especially these giant ones) have the appearance of coherence needs some explaining. (There are some good mathy explanations.) And whatever the generated words mean (if anything) will intrinsically be different from even an identical text produced by a human, if authorship matters, which in general it does. Toddlers find this out as they learn to use pronouns. Some adults become too used to this and forget. So I have to read AI text differently than human text. It's intrinsic. Sometimes the person who wrote it matters. Like my friend's 14 year-old son who sent me his analysis of some story he read with the caveat "that it's not very good." I'm like, "If I wanted good criticism, I'd go find whoever wrote about this book in the New Yorker. I'm sure it would be interesting, but today I want to know your opinion." Use Declarations are helpful. Sometimes the what of the text matters a lot more than who wrote it, most any person could have. But LLMs as-such are not persons. Their capacities and limitations are different and so for many what-centric texts, they are not a reasonable substitute. The obvious signs are redundant word use, stilted beginnings, middles, and ends. Those signs are annoying but beside the point. The issue is that you'll want to read it differently.
💡 1
👆 2
If a human person wrote a page, you know they had to type that out. And for most people that takes effort such that the existence of the page is proof that the person found it worth the effort for whatever reason.
head shaking vertically 1
Could be a bad reason. The text may not be worth the paper its written on, the wires that transmitted it, but the human person had to put in the effort. LLMs generate text for free in comparison — and that's if you factor in all of the material costs. (Roughly, there's math to be done there and incentives to fake it one way or the other.) So this is main signaling problem with using them for writing. Humans pretty much intrinsically care and machines don't. Even when someone phones-in their worst work, they had to built up enough motivation to do it.
But friends, the reactionary stance misses out on all the ways that LLMs, AI, computers, and other machines are super valuable for knowledge work. Story time. I recently graded eight papers for a "Digital Minds Sprint" some hackathon. Most papers were prompting various LLMs in various ways to you know, see how they work. All of the submissions I read had significant AI use in their text. This was absolutely acceptable and for my part super annoying because Claude is bad at technical writing: lists stupid details without making a point. So obviously I didn't read these texts like I would if a person sat down and wrote them. I don't owe Claude the respect I would give to any human person because I know basic Claude is not respectable when it come to writing. I really would have preferred each person write one human page than ten looking for the useful bits. Anyhow, getting long. What was super-helpful though was having my agent read all of their code, statistics, and analysis so that I could draw conclusions that the authors were to inexperienced to find in their own work. (In a sense this is like grading middle school work. That's fine.) I would have never never bothered to look over the data myself to double check methods and results.
To sum up, there were two good papers (two more than I thought I would find), and one was machine translated from Portuguese. All I'm saying is that AI changes the writing equation for worse and potentially better, but we have to really think about the way we use words because old defaults are off. Quick example, with Rao here, just asked my agent...
Curious. Friends are reading over contraptions.venkateshrao.com/about. I want to know if this or any of his linked pages talk about stuff from this Dreyfus "care/coping" perspective we've been working over this last week.
A quick sniff suggested Rao might have interesting things to say, this prompt amounts to deep semantic search. Much better than key-wording off of "Dreyfus" or "Heidegger" or whatever. And beyond basic LLM use, this "perspective we've been working over this last week" points to a specific, situated constellation of ideas 1,000x more specific than Google at its height would have been able to help with.
💡 1
c
many many words I pick some and then i add some, heidegger rings a bell. 🛎️ Miyazaki points to a balance, distinguishing between - bringing forth and challenging forth It may seem at times like blindfolded nitpicking but then again we are humans so there are many ways to make sense of it
i took the advantage of reading heidegger in original german which is at his level of using words still almost useless
😂 2
w
I just posted in #C050QK4917D, a talk in which I carefully, very carefully use a ukulele to channel the insights while keeping the man himself well and dead. As for Miyazaki, if you haven't looked into the full context of the "insult to life itself" quote, you may be in for a treat. The whole episode together with Miyazaki's later work: I'm thinking of "Boro the Caterpillar" and "The Boy and the Heron" specifically, tell much more nuanced story than last year's news slop.
👍 1
k
@curious_reader A few years ago I found myself in a train sitting next to a Spanish philosophy student in France who was trying to read a piece by Heidegger in German with the help of a dictionary (the paper-based variety). We chatted a bit and when he learned that I am German, he asked me if I could translate "Hineingehaltenheit" into French. Quite a challenge!
w
I was reading Dreyfus last week, who one friend describes as "America's Heidegger." I was reading, "Why Heideggerian AI Failed and How Fixing It Would Require Making It More Heideggerian," and it's pretty readable though "non-representational" confused my AI into thinking "without representations," so maybe not too clear.
c
Stellen wir das Gestell!
But 🤔
our joking points to something which is heidegger (and others too) force you to deeply think about a single word. what does it mean ? where does it connect ? In that sense it's a kind of cautionary exercise of how to read LLM prompts. The thing is it was SO difficult for me to read thsr heidegger he was nice enough to invent mostly new words and so conflating became less a thing. But AI uses our words but can contain surprises in the space between the words. In that sense reading Heidegger is a good exercise, but it is also a bit frightening to think about how difficult it is too read ( heidegger) and how much LLM output is produced everyday ( its probably a number of economics growth by now) and ahow carelessly people then accociate with it.
But I guess confronting carelessness would be in the way of accellerationism? goodreads.com/book/show/223436601-careless-people
Open in Slack
PreviousNext
Powered by