Hi Zvi, the "everyone in the UAE get's ChatGPT plus for free" is unfortunately an LLM slop hoax. This has been a partiularly redpilling one as I've seen a lot of people caught out (including me, briefly). An unfortunate tell-tale is that it seems to be exclusively Indian online tabloids reporting. Businesstoday.in (who originally broke the story) seem to have the worst possible combination of bullshit human written articles and AI slop, if you peruse the rest of their site. Here is a local news source on the matter:
Edit: I've noticed this "news story" has now been picked up by supposedly more reputable news sources, like India Today that you linked to or even the Times of India. It is telling though that only Indian news sources and twitter seem to have picked this story up at all. If I were speculating I'd suggest that there is almost certainly a link between the UAE's large Indian population and the source of the news. Clickbait but also possibly a Whatsapp-fake news migration pull factor storm. Weird anyway.
How do you think we can increase support for student visas if the entry level job market tightens? Trump could easily say something like "We need to make sure these scarce jobs are reserved for US citizen graduates, such as your children and grandchildren who are increasingly moving back home after college because they can't find a good job." Unless you straight up get rid of OPT I don't see how you thread the needle here.
Regarding the idea that we will never run out of stuff for software engineers to do, I agree with that as far as it goes. However, even without AI, probably >50% of new products fail and new features go without much adoption or are just off target in some way. If software productivity increases 5X, do we just get 5X more non-useful code? Probably even more, because the ability to come up with and vet product ideas is likely to be more constrained than the ability to implement. So at some point we need to get better at figuring out what to build so that we don't drown in an ocean of crappy features. Similar dynamic with content; we're going to need better filters.
Arguing that the climate impact of AI is negligible compared to the humans only makes sense if you intend to eliminate the humans, since otherwise their climate impact is a sunk cost. We need to eat whether or not we keep our jobs. I assume that's not the argument you intended to make.
It should be obvious that humans using AI incur additional energy and therefore climate cost over vanilla humans. I don't think that cost is currently significant but you shouldn't make it disappear with one weird accounting trick.
Well, in theory/hopefully the humans go on to do other things, and so you blame their climate impact on those other things (that AI was not able to do more efficiently). If there are no other things for humans to do, then climate may be the least of your worries?
Sam Altman: i think we should stop arguing about what year AGI will arrive and start arguing about what year the first self-replicating spaceship will take off.
I took this not as a serious question, but rather a potshot at Elon
Move the knight to put them in check. Whichever of the three possible moves they make to get out of it, you can then move the queen to put them in checkmate.
"I continue to think that a mega subscription is The Way for human viewing. Rather than pay per view, which feels bad, you pay for viewing in general, then the views are incremented, and the money is distributed based on who was viewed."
I've been a YouTube Red/Premium subscriber since basically day one. I know you go on and on about how Google can't market their way out of a used kleenex dropped in the street on a rainy day - This is Yet Another Example of that principle.
YouTube Premium is the mega subscription model you're describing, and it already works brilliantly - Google just can't market it to save their lives.
Nobody and I mean NOBODY (except The Spiffing Brit in one throwaway line in one video¹) realizes that YouTube splits Premium subscription revenue 55/45 with creators, just like ad revenue. But here's the kicker: it fundamentally changes the entire ecosystem. Instead of viewers being the product sold to advertisers, we become the customers. Creators get paid MORE per Premium view than ad views, without worrying about "advertiser-friendly" content guidelines. I never get that "feels bad" moment of deciding if a video is worth X cents - I just watch what interests me, guilt-free.
The model removes ALL the friction: no ads, no ad-blocker wars, creators get stable income, and viewers can support creators without thinking about it. YouTube Premium is proof that the mega subscription model works. The only failure is that after nearly a decade, Google still hasn't figured out how to explain this to people. They've positioned it as "YouTube without ads" when it's actually "become a patron of every creator you watch, automatically."
¹ Explained briefly from 1:24 to 2:40 in "YouTube Premium Is Broken." by The Spiffing Brit, 14 August 2021. "Most creators see basically no revenue from Premium each month."
Elon Musk clearly follows AI Safety Memes. He's retweeted and commented on the account a bunch of times. Not sure if he officially "follows" the account, but AISM is definitely in his feed.
As I mentioned in a comment on a previous post which mentioned AI improv scenes, I am an improv performer and instructor in addition to my day job. Claude 4 opus is the first time I was able to get a solid scene, but it took some iteration.
Some of this is that a lot of the writing about improvizing is aggressively mediocre, so when you prompt the model that it's doing an improv scene it indexes strongly on that mediocre training set. Some of it is also taste, in the middle of a scene good performers are evaluating all the offers, following the strong/interesting ones, and setting the weaker ones aside. Below is a transcript of the best scene I was able to do with Opus.
Actor 1: [Walks out on stage, begins digging through cupboards, rifling through them as if looking for something. Shouts offstage] It's not in the pantry.
Actor 2: [Rushes in from stage left, carrying an armful of pool noodles] Well it's definitely not in the garage either! Just these old pool noodles and that box of your mom's wedding china we promised we'd never touch. [Drops the pool noodles with a clatter] Honey, are you SURE you put it back after the Henderson's barbecue? Because I distinctly remember you saying you'd "put it somewhere logical this time." [Starts checking jacket pockets hanging by the door]
Actor 1: [Crosses over and hugs Actor 2, then pats the sides of their shoulders in a soothing but condescending manner] Respectfully honey, this is our last piece of cursed gold, do you really think that I would be so reckless as to lose it after showing the Hendersons. To imagine you think so little of me shocks rocks me to my core. Now come, stop distracting me. [Digs around a floor level cupboard stage left until finding something. Says "Aha", stands up, affixing a headlamp to head] Now I'm equipped to look properly. Remind me to reorganize that one, I keep meaning to get all that junk sorted but for whatever reason I never make the time.
Actor 2: [Watches with mixture of affection and exasperation as Actor 1 fumbles with headlamp] Oh yes, you'll reorganize it. Just like you reorganized the attic three years ago, which is why we still can't find the Christmas decorations. [Picks up one of the pool noodles, peers inside it hopefully] And for the record, I don't think you LOST it. I think you put it somewhere "clever" again. Like when you hid the car keys in the freezer because "thieves never check there." [Suddenly freezes] Wait. The Hendersons. Didn't their daughter pick up your jacket by mistake when they were leaving? The jacket with all the inside pockets you had specially made for "important items"? [Turns to face Actor 1, headlamp beam hitting eyes] Ow! Could you— never mind. Should we perhaps call them before the curse, you know... activates? How long do we have again?
Actor 1: [Flips headlamp up so it's not in Actor 2's eyes, eyes drifting up in thought] Well if the Henderson girl took the gold, even without meaning to, the demon would interpret that as her holding covetousness in her heart, and thus as an invitation to our plane. She would then of course have to pass the trials. [Walks across room, opens closet, starts donning wizard battle garb] They're more your friends than mine, what's the etiquette here. And what do you think of Cassie, will she prove worthy of crossing the threshold? Also, do you think we need to bring a bottle of wine? [Begins to wrap Actor 2's sacred combat vestments around them]
Actor 2: [Lifts arms obediently as Actor 1 wraps the vestments, sighing] Cassie Henderson is eleven years old and still sleeps with a nightlight. Last week she cried because her hamster looked at her "weird." I highly doubt she's ready for the Threshold of Eternal Judgment. [Adjusts the vestments, checking various hidden pockets] Though I suppose her mother did survive that whole incident with the haunted Pinterest board, so maybe it's genetic? [Grabs ceremonial staff from closet, tests its weight] And yes, definitely bring wine. The Hendersons are "bring a dish to share" people even during normal visits. Showing up to save their daughter from demonic trials empty-handed would be... tacky. [Pauses, looking at Actor 1 with sudden worry] Oh god, what if they're eating dinner? Do we wait for them to finish? I've never had to do an emergency exorcism on people I play bunco with.
Bank teller jobs are a result of Too Big To Fail regulation decisions that squeeze out the number of regional banks and limit small banks for the big guys, I think. Not technology but consolidation.
I have a feeling that for Claude to be popular among general audience they have to solve those punishing message limits. Enthusiasts are eager to work around them, but it can be frustrating for everyone else, I guess.
> We can agree that one key such objection, which he calls the ‘capability denialist’ (a term I intend to steal) is essentially refuted now, and he says we hear about it less and less. Alas, this continues to be the most common objection, that the AI won’t be capable enough to worry about, although this is often framed very differently than that, such as saying ‘it will only be a tool.’
I'd propose distinguishing between 'capability denialism' and 'agency denialism', with a reasonable argument that the latter could be seen as a special case of the former. I think there's a qualitative difference between statements of the form "AI will not be able to do [x]" and those of the form "AI will not do [x] unless someone (deliberately?) instructs it to do [x]".
Learn to fear the automated context repack. Opus 4 is pretty excellent at writing code and I've got it doing a lot more work independently, but every claude code automatic context repack is a chance for things to go off the rails. I've had it forget how to use linux all of the sudden and have it suddenly start to invoke dotnet commands or whatever. Bridging that gap seems fraught for automation. Anyone else run into this?
You linked the same SMBC twice.
Hi Zvi, the "everyone in the UAE get's ChatGPT plus for free" is unfortunately an LLM slop hoax. This has been a partiularly redpilling one as I've seen a lot of people caught out (including me, briefly). An unfortunate tell-tale is that it seems to be exclusively Indian online tabloids reporting. Businesstoday.in (who originally broke the story) seem to have the worst possible combination of bullshit human written articles and AI slop, if you peruse the rest of their site. Here is a local news source on the matter:
https://gulfnews.com/business/markets/is-openai-giving-chatgpt-for-free-to-all-uae-residents-know-whats-true-1.500141497
Edit: I've noticed this "news story" has now been picked up by supposedly more reputable news sources, like India Today that you linked to or even the Times of India. It is telling though that only Indian news sources and twitter seem to have picked this story up at all. If I were speculating I'd suggest that there is almost certainly a link between the UAE's large Indian population and the source of the news. Clickbait but also possibly a Whatsapp-fake news migration pull factor storm. Weird anyway.
Huh, okay, I will put in a correction next week (putting it in now would do nothing)
Podcast episode for this post:
https://open.substack.com/pub/dwatvpodcast/p/ai-118-claude-ascendant?r=67y1h&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
Zvi, how do you maintain this level of output? In the Tyler sense, what is your productivity function?
This newsletter is priceless and the amount you crank out is just wild to me.
How do you think we can increase support for student visas if the entry level job market tightens? Trump could easily say something like "We need to make sure these scarce jobs are reserved for US citizen graduates, such as your children and grandchildren who are increasingly moving back home after college because they can't find a good job." Unless you straight up get rid of OPT I don't see how you thread the needle here.
Regarding the idea that we will never run out of stuff for software engineers to do, I agree with that as far as it goes. However, even without AI, probably >50% of new products fail and new features go without much adoption or are just off target in some way. If software productivity increases 5X, do we just get 5X more non-useful code? Probably even more, because the ability to come up with and vet product ideas is likely to be more constrained than the ability to implement. So at some point we need to get better at figuring out what to build so that we don't drown in an ocean of crappy features. Similar dynamic with content; we're going to need better filters.
Arguing that the climate impact of AI is negligible compared to the humans only makes sense if you intend to eliminate the humans, since otherwise their climate impact is a sunk cost. We need to eat whether or not we keep our jobs. I assume that's not the argument you intended to make.
It should be obvious that humans using AI incur additional energy and therefore climate cost over vanilla humans. I don't think that cost is currently significant but you shouldn't make it disappear with one weird accounting trick.
Well, in theory/hopefully the humans go on to do other things, and so you blame their climate impact on those other things (that AI was not able to do more efficiently). If there are no other things for humans to do, then climate may be the least of your worries?
Sam Altman: i think we should stop arguing about what year AGI will arrive and start arguing about what year the first self-replicating spaceship will take off.
I took this not as a serious question, but rather a potshot at Elon
What's the best move for the chess game?
Move the knight to put them in check. Whichever of the three possible moves they make to get out of it, you can then move the queen to put them in checkmate.
I believe it’s (rot13’ed)
Xavtug gb Q3, gura nal bs juvgr (n) zbivat gur xvat, (o) gnxvat gur xavtug jvgu nabgure xavtug, be (p) gnxvat gur xavtug jvgu gur dhrra erfhyg va purpxzngr arkg ghea sebz Oynpx'f dhrra.
Gur "gevpx" vf gung vg erdhverf n guerr-cebatrq gerr frnepu vafgrnq bs pbafgenvavat juvgr gb n fvatyr frg bs ybfvat zbirf.
"I continue to think that a mega subscription is The Way for human viewing. Rather than pay per view, which feels bad, you pay for viewing in general, then the views are incremented, and the money is distributed based on who was viewed."
I've been a YouTube Red/Premium subscriber since basically day one. I know you go on and on about how Google can't market their way out of a used kleenex dropped in the street on a rainy day - This is Yet Another Example of that principle.
YouTube Premium is the mega subscription model you're describing, and it already works brilliantly - Google just can't market it to save their lives.
Nobody and I mean NOBODY (except The Spiffing Brit in one throwaway line in one video¹) realizes that YouTube splits Premium subscription revenue 55/45 with creators, just like ad revenue. But here's the kicker: it fundamentally changes the entire ecosystem. Instead of viewers being the product sold to advertisers, we become the customers. Creators get paid MORE per Premium view than ad views, without worrying about "advertiser-friendly" content guidelines. I never get that "feels bad" moment of deciding if a video is worth X cents - I just watch what interests me, guilt-free.
The model removes ALL the friction: no ads, no ad-blocker wars, creators get stable income, and viewers can support creators without thinking about it. YouTube Premium is proof that the mega subscription model works. The only failure is that after nearly a decade, Google still hasn't figured out how to explain this to people. They've positioned it as "YouTube without ads" when it's actually "become a patron of every creator you watch, automatically."
¹ Explained briefly from 1:24 to 2:40 in "YouTube Premium Is Broken." by The Spiffing Brit, 14 August 2021. "Most creators see basically no revenue from Premium each month."
Elon Musk clearly follows AI Safety Memes. He's retweeted and commented on the account a bunch of times. Not sure if he officially "follows" the account, but AISM is definitely in his feed.
As I mentioned in a comment on a previous post which mentioned AI improv scenes, I am an improv performer and instructor in addition to my day job. Claude 4 opus is the first time I was able to get a solid scene, but it took some iteration.
Some of this is that a lot of the writing about improvizing is aggressively mediocre, so when you prompt the model that it's doing an improv scene it indexes strongly on that mediocre training set. Some of it is also taste, in the middle of a scene good performers are evaluating all the offers, following the strong/interesting ones, and setting the weaker ones aside. Below is a transcript of the best scene I was able to do with Opus.
Actor 1: [Walks out on stage, begins digging through cupboards, rifling through them as if looking for something. Shouts offstage] It's not in the pantry.
Actor 2: [Rushes in from stage left, carrying an armful of pool noodles] Well it's definitely not in the garage either! Just these old pool noodles and that box of your mom's wedding china we promised we'd never touch. [Drops the pool noodles with a clatter] Honey, are you SURE you put it back after the Henderson's barbecue? Because I distinctly remember you saying you'd "put it somewhere logical this time." [Starts checking jacket pockets hanging by the door]
Actor 1: [Crosses over and hugs Actor 2, then pats the sides of their shoulders in a soothing but condescending manner] Respectfully honey, this is our last piece of cursed gold, do you really think that I would be so reckless as to lose it after showing the Hendersons. To imagine you think so little of me shocks rocks me to my core. Now come, stop distracting me. [Digs around a floor level cupboard stage left until finding something. Says "Aha", stands up, affixing a headlamp to head] Now I'm equipped to look properly. Remind me to reorganize that one, I keep meaning to get all that junk sorted but for whatever reason I never make the time.
Actor 2: [Watches with mixture of affection and exasperation as Actor 1 fumbles with headlamp] Oh yes, you'll reorganize it. Just like you reorganized the attic three years ago, which is why we still can't find the Christmas decorations. [Picks up one of the pool noodles, peers inside it hopefully] And for the record, I don't think you LOST it. I think you put it somewhere "clever" again. Like when you hid the car keys in the freezer because "thieves never check there." [Suddenly freezes] Wait. The Hendersons. Didn't their daughter pick up your jacket by mistake when they were leaving? The jacket with all the inside pockets you had specially made for "important items"? [Turns to face Actor 1, headlamp beam hitting eyes] Ow! Could you— never mind. Should we perhaps call them before the curse, you know... activates? How long do we have again?
Actor 1: [Flips headlamp up so it's not in Actor 2's eyes, eyes drifting up in thought] Well if the Henderson girl took the gold, even without meaning to, the demon would interpret that as her holding covetousness in her heart, and thus as an invitation to our plane. She would then of course have to pass the trials. [Walks across room, opens closet, starts donning wizard battle garb] They're more your friends than mine, what's the etiquette here. And what do you think of Cassie, will she prove worthy of crossing the threshold? Also, do you think we need to bring a bottle of wine? [Begins to wrap Actor 2's sacred combat vestments around them]
Actor 2: [Lifts arms obediently as Actor 1 wraps the vestments, sighing] Cassie Henderson is eleven years old and still sleeps with a nightlight. Last week she cried because her hamster looked at her "weird." I highly doubt she's ready for the Threshold of Eternal Judgment. [Adjusts the vestments, checking various hidden pockets] Though I suppose her mother did survive that whole incident with the haunted Pinterest board, so maybe it's genetic? [Grabs ceremonial staff from closet, tests its weight] And yes, definitely bring wine. The Hendersons are "bring a dish to share" people even during normal visits. Showing up to save their daughter from demonic trials empty-handed would be... tacky. [Pauses, looking at Actor 1 with sudden worry] Oh god, what if they're eating dinner? Do we wait for them to finish? I've never had to do an emergency exorcism on people I play bunco with.
[Lights fade to black]
Bank teller jobs are a result of Too Big To Fail regulation decisions that squeeze out the number of regional banks and limit small banks for the big guys, I think. Not technology but consolidation.
I have a feeling that for Claude to be popular among general audience they have to solve those punishing message limits. Enthusiasts are eager to work around them, but it can be frustrating for everyone else, I guess.
> We can agree that one key such objection, which he calls the ‘capability denialist’ (a term I intend to steal) is essentially refuted now, and he says we hear about it less and less. Alas, this continues to be the most common objection, that the AI won’t be capable enough to worry about, although this is often framed very differently than that, such as saying ‘it will only be a tool.’
I'd propose distinguishing between 'capability denialism' and 'agency denialism', with a reasonable argument that the latter could be seen as a special case of the former. I think there's a qualitative difference between statements of the form "AI will not be able to do [x]" and those of the form "AI will not do [x] unless someone (deliberately?) instructs it to do [x]".
Learn to fear the automated context repack. Opus 4 is pretty excellent at writing code and I've got it doing a lot more work independently, but every claude code automatic context repack is a chance for things to go off the rails. I've had it forget how to use linux all of the sudden and have it suddenly start to invoke dotnet commands or whatever. Bridging that gap seems fraught for automation. Anyone else run into this?