Eliezer Yudkowsky Episode is Out!
Added 2023-06-10 04:57:28 +0000 UTC
It is out! This one took quite a while to edit as you might imagine. It is still very long but you should be used to that by now.
In any case, interested in any and all feedback.
Sorry for no advance release, we released it as soon as we finished editing.
The bonus Gurometer episode is coming out today though!
Comments
I discussed this episode with chat GPT-4. I now know what it’s like to be pitied by a non-sentient intelligence.
Paul Ditta
2025-01-22 10:18:26 +0000 UTCThat analogy was really cringe-worthy. I cannot fathom the patience it requires to listen to all that stuff you're decoding. In case you'd like to hear a more convincing story (in my opinion) about AI risks, I suggest Forrest Landry. He discussed the topic on the Jim Rutt Show, episodes 181 and 183.
Roland Weber
2023-06-25 21:48:14 +0000 UTCbillw2011, I share your view that if there's even a small chance that AI is an extinction risk, it should be taken very seriously. Yudkowsky's claim is stronger though--he expects AI is going to kill us all. It matters for response what the likelihood is. If extinction by AI is very likely to happen (say 90% chance), that to me means drop everything and work on nothing but solving the problem. If there's a small chance of extinction by AI (say 0.01% chance), to me it would warrant urgent study to clarify and/or reduce the risk.
Daniel
2023-06-17 05:07:23 +0000 UTCI’m a bit disappointed that you failed to mention the amazing fact that he wrote a 600k+ word Harry Potter fanfic, which apparently is critically acclaimed
Josh
2023-06-16 12:25:58 +0000 UTCI can't believe I've finally heard someone whose mayonnaise voice is more repulsive to me than that of Ira Glass. And with the added stabs of smug self-satisfaction.
JustTheWorstEver
2023-06-15 18:44:13 +0000 UTCIt’s true. Matt has a more gestalt memory of the content. He usually doesn’t know the clips in advance.
Christopher Kavanagh
2023-06-14 07:58:31 +0000 UTCOh nice! That’s cool.
Christopher Kavanagh
2023-06-14 07:57:50 +0000 UTCI'm a bit late to this one but the phrase from Blade Runner you were struggling to recall was "like tears in the rain" :D here's the full thing: "I've seen things you people wouldn't believe... Attack ships on fire off the shoulder of Orion... I watched C-beams glitter in the dark near the Tannhäuser Gate. All those moments will be lost in time, like tears in rain..." - spoken by Roy Batty the android right before he died
ketracel-white
2023-06-13 21:04:33 +0000 UTCI have a pretty interesting Eliezer Yudkowsky connection. So, for the past five years, I've been working on my Sci-fi webcomic called "Seed." It just wrapped up last month with its 150th episode. It has been my attempt to tell a grounded story about AI, and I would be thrilled if you guys check it out. I suspect Matt will really enjoy it. Anyway, here's the Yudkowsky part. It turns out the term "Seed AI" was first coined by Yudkowsky, and to my surprise, one day I saw him in my comments section. He commented pretty positively on like ten episodes. So I've got to give it to him; the man obviously has pretty good taste in sci-fi. 😂 So yeah, If he uses his right to reply, which I'm pretty sure he will for some reason, consider asking him about "Seed."
Said Polat
2023-06-13 20:11:23 +0000 UTCI love the idea that Matt has genuine questions almost as if he's listening to the content for the first time, and Chris immediately queues up relevant audio clips. If this isn't the case, please never shatter my illusion. ❤️
Jake Lawrence
2023-06-13 14:50:32 +0000 UTCI used 3.5 too, and no clarification was needed - it gave the above as the first answer. And all in one reply it asked for additional details, like dimensions, sizes, so it was ready to refine the answer. And yes, perhaps such question was in its learning material, but hey - my students do the same type of regurgitation during exams! At any rate I am curious how many years ago Matt did his test ;)
Tomasz
2023-06-13 14:40:36 +0000 UTCGreat job gentlemen, I could not stop laughing this episode when the discussion went to "Lex made of code who is as smart as von Neumann, is trapped inside a box made by aliens who are like trees and have head bopping factories"
MaxPlan
2023-06-13 05:48:02 +0000 UTC😂 I'm waiting for Matt to say something like "look it's a free country, and if people want to do philosophy in the privacy of their own homes that's FINE, I just want them to stop SHOVING IT IN MY FACE when I'm INNOCENTLY going about my intellectual business like a NORMAL academic" "AI won't kill us all... just the philosophers, AND THEY DON'T COUNT" But where's Chris in all this? Are cognitive anthropologist sufficiently scientific to be allowed to live?
Artemis Green
2023-06-13 03:59:55 +0000 UTCDamn, dare I say that Lex almost sounded reasonable and nuanced sometimes. In comparison to Yudkowsky at least. Kinda like how Bush seems like a halfway intelligent and decent person in comparison to Trump.
Cesare, But Def Not A Borgia
2023-06-12 22:40:49 +0000 UTCYudkowsky relies so heavily on how bad it would be if he were right as compared to actual evidence that his predictions are going to come true. It's a very emotion based argument, ironically. If you repeat "we all die" enough times, yeah it starts to sound scary.
Ben Godek
2023-06-12 17:03:41 +0000 UTC"I don't hate poetry, or people who like poetry, and so long as they admit that it DOESN'T MEAN ANYTHING, that it's just a STUPID WASTE OF TIME, and start studying MATTHEMATICS and stop using fake words that AREN'T EVEN REAL, nobody's family has to die a horrible death." -Matthew Browne, 2026, giving orders to his philosopher-hunting drone / GPT5-powered best friend ("beats Twitter!")
Exai
2023-06-12 14:05:10 +0000 UTCI was a bit surprised at how obvious the flaws in Yudkowsky's arguments were - even to this luddite. So I was mostly here for the banter. I'm still trying to work out why Matt is so irked by continental philosophy, Dostoyevsky, and people who never grow out of searching their souls about what it means to be a moral human :P Edit: A fedora?! PLEASE EXPLAIN.
Artemis Green
2023-06-12 13:52:04 +0000 UTCEven GPT 3.5 is getting this correct for me, once I clarified I meant edge intersections. But this seems like it would be a reasonably common test question anyway, so could be answered by regurgitation rather than reasoning.
billw2011
2023-06-12 08:24:37 +0000 UTCI really don't understand this, maybe someone can explain. Why would he need to show that "we all die" is *most* likely?! If "we all die" is even remotely possible we should be taking it very seriously. It seems like a basic misunderstanding of the game theory, but I find it hard to see how people can miss something so obvious. Even a 1% chance we all die is really really terrible odds, and I haven't really heard any of the Pollyanna's directly address this, and how it specifically should apply to reasoning about AI threat.
billw2011
2023-06-12 08:01:57 +0000 UTCHow long ago did you check that square/circle problem, Matt? Because today's GPT says: The maximal number of intersection points between a square and a circle in a plane depends on the relative sizes and positions of the two shapes. In general, a square can intersect a circle at a maximum of 8 points. This occurs when the circle is located in such a way that its boundary touches or crosses the sides of the square at four points, and additionally intersects the corners of the square at four points. (It then goes on to list other factors and asks for more details.) So, are we doomed yet? ;)
Tomasz
2023-06-11 22:16:54 +0000 UTCI do this when I'm talking about something really tricky or complex - or I will break eye contact and look away for a bit while I talk. I mean, I guess you could call it nerdy and awkward, but for me I'm just blocking out a distraction (someone's face) so that I can focus on thinking through the issues and choosing my words. I've seen judges hear cases mostly with their eyes closed or staring at the ceiling and you think they might be off with the fairies until they open their mouths and what they say shows they've been listening intently the whole time. I think it's just a reflection of the way some people process information.
Artemis Green
2023-06-11 01:22:03 +0000 UTCStill listening but great work. This is not a criticism of Lex and Eliezer but did you notice that both close their eyes when speaking regularly? Lex does it regularly is one thing, but seeing both doing it was interesting. It makes me wonder how much of their views comes from their nerdy awkward personalities. It also makes me self conscious, I would probably do something like that without thinking if I was being interviewed.
MaxPlan
2023-06-11 00:00:36 +0000 UTCI think Yudkowsky needs to give a better argument for why "we all die" is the most likely consequence of AI superintelligence. I searched around to see if he's made one publicly but didn't have any luck. I might dare to ask him directly. Matt may be right that he's relying a lot on intuitions developed through sci-fi and projection. By the way, I could totally see this Rogan conspiracy theory about advanced AI manipulating the world catching on, perhaps with elites at the controls.
Daniel
2023-06-10 23:29:28 +0000 UTC"We're just feeble biological automata; we're absolutely powerless in the face of large matrices getting multiplied with each other." Not just any matrices, I would add: these are matrices with FLOATING POINT numbers.
Maarten Wesselius
2023-06-10 11:23:19 +0000 UTC3hrs 12mins. Rumour has it even Hiroo Onoda tapped out half way through listening to this episode (5 stars though)
NIALL DAVISON
2023-06-10 07:49:52 +0000 UTC
