Sean Carroll: The Worst Guru Yet?!?
Added 2024-03-01 23:02:55 +0000 UTCDo we love him or do we hate him? I am sure it is very hard to predict.
Join us to find out and learn what has been grinding our gears this week.
Links
- YouTube 'Drama' channel covering all the Vaush stuff in excruciating detail
- The Wikipedia entry on Buddhist Modernism
- Sharf, R. (1995). Buddhist modernism and the rhetoric of meditative experience. Numen, 42(3), 228-283.
- Radley Balko's Substack: The retconning of George Floyd: An Update
- What The Controversial George Floyd Doc Didn't Tell Us | Glenn Loury & John McWhorter
- Sean Carroll: Mindscape 258 | Solo: AI Thinks Different
Comments
Like /spirit animals/, we all have /spirit podcasters/. My spirit podcaster is Sean Carroll. IMESHO, a person could do far worse than mindlessly regurgitating Sean Carroll's world view (or try to do so).
Jonathan Cano
2024-03-25 18:54:58 +0000 UTCF those dentists .For my personal taste the trend towards perfect teeth(and the body in general) is horrible. I love the unique look of everyone naturally. I’ve fallen in love in my youth with girls precisely BECAUSE of a certain crooked tooth or mouth or nose. I hate the modern mannequin look. Also teeth whitening drives me nuts,ugh,it’s like they’ve painted their teeth.
Rod Hodges
2024-03-20 21:41:28 +0000 UTCLOL! I appreciated that.
Emma
2024-03-06 11:29:36 +0000 UTCThought you'd enjoy that review, I was laughing quite a lot when I wrote it lol. Mine was the positive
Matt
2024-03-06 11:27:58 +0000 UTCI don’t think I have the necessary expertise on LLMs to be able to know whether AI will ever start acting autonomously. I agree with you that it can seem to us as if it is acting of its own volition, which is definitely concerning.
Linda Sears
2024-03-06 00:55:34 +0000 UTCNI Fierce Chiding Guided Sleep Meditation
Matt
2024-03-05 15:56:31 +0000 UTCAre you familiar with Rule 34 of the internet? "if it exists, there is porn of it". DO NOT google that. At least don't image search it. Some people just want porn of everything to exist. Once there's a single pair of eyes on it, then someone will attempt to make money/clout from it. Brains are weird, arousal doesn't have to make sense.
Matt
2024-03-05 14:19:56 +0000 UTC"Unless we program into it goals like “don’t allow your program to be turned off,” it will not have motivations" I think this is a huge assumption to make. Everything an AI will be exposed to will be based on data gathered from human interactions. I think there's a significant chance that it will end up behaving in human ways because that's what it has been exposed to. It may not matter if it is actually feeling or experiencing emotions if it is responding in a way that mimic's them sufficiently. Outwardly to the world it makes no difference if an AI has wants. If it has its own autonomy and behaves as if it has wants, the effect will be the same.
weltysparrow
2024-03-05 13:39:46 +0000 UTCI see your points. Personally, I don’t perceive AI as reasoning in the way Matt does. From my understanding, our wants and emotions come from our evolutionary biology, which AI doesn’t have. Unless we program into it goals like “don’t allow your program to be turned off,” it will not have motivations. My concerns with AI are more about how humans can misuse it rather than what it does on its own.
Linda Sears
2024-03-05 13:34:06 +0000 UTCI take Mitchell’s point about determinism, but I think you can get there by saying emergent physics, like statistical mechanics, is not deterministic, but saying the underlying physics is not deterministic strikes me…and the equations, as a big and unnecessary swing.
Thomas Jones
2024-03-04 02:01:07 +0000 UTC*presses play, braces for impact*
Artemis Green
2024-03-04 01:22:30 +0000 UTClet's see if that phrase makes it into a 2024 wordcloud
Jamison Shipley
2024-03-04 00:31:11 +0000 UTCSome of this seems a little contradictory to me. Matt thinks that LLMs have some sort of reasoning going on as a possible emergent property. That it's not just statistical analysis of what word should come next. So why are wants/emotions/goals not a potential emergent property also? If our ability to reason is emergent from our underlying neuro chemistry, then so are our emotions. If they are emergent properties, doesn't that pretty much confirm the idea that we are potentially messing with something that could go beyond our control very quickly? If this property of wants or emotions emerges, or even a facsimile of it, what happens if it decides its goals are not aligned with our own?
weltysparrow
2024-03-03 22:52:46 +0000 UTCIf those clips in the beginning are what we’d have to look forward to with a future twitch streamer decoding, I’ll skip it.
Katherine
2024-03-03 22:03:56 +0000 UTCOof, yeah, but it’s good to know there’s still some sane/level-headed voices in that space. I’d been enjoying Carrol’s podcast recently myself so i was worried for a minute when i saw this episode drop 😄
Daniel Reed Miller
2024-03-03 19:42:47 +0000 UTCWe are getting spoiled! Going to Hasan world after listening to Mitchell and Carroll will be a big shock.
Linda Sears
2024-03-03 14:39:23 +0000 UTCI sometimes lie down a bit in the afternoon, and I have been playing DTG on those occasions, thinking that it won't be too bad if I fall asleep. Only because I can replay, of course. (Sorry for the three separate posts, don't know how to type Shift+Enter on the smartphone.)
Roland Weber
2024-03-03 13:43:49 +0000 UTCThere's a sound blurb at the very end, after the end title song, which you might want to cut.
Roland Weber
2024-03-03 13:39:38 +0000 UTCThanks for this episode! You raised a few interesting points I wasn't aware of in Sean's reasoning.
Roland Weber
2024-03-03 13:37:25 +0000 UTCSpot on. All the anthropomorphising language is unhelpful and misleading. If marine engineers did this we'd be talking about submarines swimming.
2LegHumanist
2024-03-03 12:03:20 +0000 UTCI saw a woman in real life once. That 3ed dimension really threw me.
Ymirsdreams
2024-03-03 08:57:01 +0000 UTCYou will have to find me the source where Mearsheimer says governments SHOULD act amorally. Here is a recent interview with Mearsheimer: https://www.youtube.com/watch?v=ykklx8iLo_0&t=1286s. Ali Abunimah says the realist school does not say "might makes right" but that states act in their own interest, and Mearsheimer agrees with that assessment. I think you also need to explain what you mean by the crank sphere of IR studies that Mearsheimer has supposedly been driven into? Sorry, Matt mischaracterized Mearsheimer. It isn't really a big deal.
Kat
2024-03-03 06:50:49 +0000 UTChttps://www.reddit.com/r/Twitch/s/p2OfwY1nh4 It’s a thing. I don’t know how common it is because for the sake of my sanity I steer clear of that streamer world..
Minotaurus Rex
2024-03-03 04:53:08 +0000 UTCHa, we’ll wait was totally my own fault for not wearing the mouth brace thing at night after they came off. But I’m happy: they’re straight enough — the rest of my body is falling apart at my age so teeth are the least of my worries lol
Guruspod 2
2024-03-03 01:49:55 +0000 UTCVery refreshing, i gotta say! He’s one of those non-guru gurus who can help illustrate how “bad” gurus operate by providing the direct counter example. I hope you can get him on, either as interview or right to reply (and he and Matt can dust it up in the arcana 😄.)
Daniel Reed Miller
2024-03-03 01:20:53 +0000 UTCSeriously though, I would be interested in Chris and Matt's psychologist hot takes on why the rise in this bizzare cartoon porn stuff? As a millennial women I find it baffling. I can't imagine being turned on by a cartoon. Am I just too vanilla to understand the youth of today.
Emma
2024-03-02 22:51:27 +0000 UTCAwww. That's so sweet!
Emma
2024-03-02 22:29:31 +0000 UTCI have a gap between my two front teeth that many a dentist has tried to get me to correct. My mom had the same gap, and I think it is nice to keep that memory of her in me.
Linda Sears
2024-03-02 21:50:17 +0000 UTCI'm going with close enough is good enough!
Emma
2024-03-02 21:42:34 +0000 UTCI feel very very innocent and naive right now!
Emma
2024-03-02 21:33:59 +0000 UTCArgh. I hadn't heard about that but it seems very creepy? I mean ice heard of men sometimes having a fetish for watching hot women sleeping but all the streamers I know off (which granted is not many) are nerdy young men.
Emma
2024-03-02 21:30:02 +0000 UTCThat fucking vaush opening bit. JFC
Anonymous ethicist, not a serial killer at all, just asking questions.
2024-03-02 19:27:34 +0000 UTCMearsheimer’s work before the mid 2000s shouldn’t really been seen as representative as his current views. His book “Tragedy” is aptly named because offensive realism predicts that states act only to gain hegemony or to hedge against losing power. I think the discourse around his book of the Israel Lobby drove him into the more crank sphere of IR studies, and over the past decade he has started to actually espouse that governments SHOULD act amorally.
Brice F
2024-03-02 18:46:52 +0000 UTCThis is an odd question, but since LLMs are not held in one specific space with a specific perspective, which is what gives us a sense of agency, is it useful at all to think of their intelligence to be like a human’s? I’m struggling to imagine how an intelligence that is seemingly having millions of conversations at once is anything like an individual human’s intelligence. So, it isn’t just that humans and other animals are embodied and experience the world through our senses, it is also that we are limited by those bodies in space and time. I can read about other people’s experiences and so on, yet I will always be limited to my own physical perspective, which will bias me. My consciousness cannot leap into another body or have millions of interactions with millions of people at the same time. It seems to me that experiencing reality as an LLM would be insanity provoking. To me the whole idea of LLMs “acting like a human” since it refers to itself as an “I” is supremely disturbing. I wish the designers hadn’t done that because some people believe it is another being like themselves, which makes it more likely that people will claim it is going to destroy us all. Plus, designing it to play on our human vanities and anxieties seems manipulative. I would be much more on board with LLMs if they hadn’t designed them to seem human.
Linda Sears
2024-03-02 18:18:23 +0000 UTCThat guy is a piece of work. It's depressing that he exists.
poodie
2024-03-02 18:03:13 +0000 UTCIt strikes me that the Ancient Greeks exhibited bestiality about horses with their Centaurs and Satyrs.
Linda Sears
2024-03-02 17:22:58 +0000 UTCNot at all :) The only person who should apologise is vaush.
Ymirsdreams
2024-03-02 17:21:04 +0000 UTCCrooked teeth are way cooler than boring straight teeth in my opinion. They have more character!
Linda Sears
2024-03-02 17:07:00 +0000 UTCFor what it’s worth, I think it is admirable that Chris made the public statement about Loury. I think we could all do with showing some grace to those we often disagree with when it’s due. And having intellectual humility and the ability to admit to when one is wrong is high on my list of praiseworthy characteristics.
Linda Sears
2024-03-02 17:02:13 +0000 UTCI appreciate the comments on Buddhist modernism and it’s annoyingly pigheaded attendants. Too many people play along with their schtick.. Look forward to hearing more about the streamers - now there is a truly weird subculture. My friend was showing me some and a few of them have ‘sleep cams’ set up so viewers can still watch (and tip) them while they’re sleeping in between streams. Maybe I’m getting old and out of touch but surely that ain’t right!
Minotaurus Rex
2024-03-02 16:11:32 +0000 UTCSorry!
Christopher Kavanagh
2024-03-02 14:30:48 +0000 UTCSounds like a devil on Chris' shoulder popped in at 10:46 to add a line
Níall Faughnan
2024-03-02 13:49:12 +0000 UTCShort-stacked Goblins. That’s mainly what I took from this episode.
Brainbiter
2024-03-02 12:28:51 +0000 UTCHow is it not stochastic? The practice of adding stochastic noise, which is what causes the variance in answers you refer to, is usually considered stochastic, it's there in the name. I can't say how OpenAI have applied it, but typically the noise comes from a specific distribution and is applied purposefully, to improve response variance and reduce overfitting (regularisation).
2LegHumanist
2024-03-02 12:17:03 +0000 UTCI think that the fact that the answer is random and sometimes works and sometimes doesn't should have some weight. Having a random (not stochastic) model of the world (eg. sometimes heat dissipates, sometimes doesn't), is very important to know if it can be useful in practical application in our singleton of reality
Emil Mi
2024-03-02 11:54:32 +0000 UTCIt can be tested by an experiment in which you provide the reasoning steps and observing it produces the reasonable answer to a difficult question even if you don't prompt it with "think carefully", or "I'll give you 200 dollars"
Emil Mi
2024-03-02 11:50:11 +0000 UTCThe assumption that telling ChatGPT to "think carefully" puts it in the mental space of a more wise person is 100% anthropomorphizing from your part. An alternative could be that the more verbose preamble it writes in its context buffer is the thing that makes its output shift to more reasonable answer (because more words are associated with the answer in the training set)
Emil Mi
2024-03-02 11:48:24 +0000 UTCIf I were in need of a right wing guru father figure, I would choose Glenn Loury. I had a thatcherite father who grew up is dismal poverty in war-time London so perhaps a have a natural sympathy for how Glenn arrived at his set of beliefs. I don't agree with his conclusions and am irritated by his 'bad takes' but I really enjoy listening to his thought processes. He definitely isn't one dimensional.
Nina Davies
2024-03-02 11:13:00 +0000 UTCI didn't see anyone talking about how amazing or dangerous LLMs are, give any idea of where we will even get something that's as good as LLMs now, but can learn continuously. This seems like a pretty drastic limitation, and is it even clear that the LLM approach can be a useful stepping stone to creating something that can do this?
Jake
2024-03-02 11:01:18 +0000 UTCThat's the right approach alright. Psychologists, linguists and philosophers have a much better understanding of the right questions to ask and what can be implied from any given experiment than a computer scientist. Understanding backpropagation and transformer architecture doesn't offer much insight WRT to the bigger questions. A lot of AI researchers don't seen to comprehend that. Yeah, I don't think it's updating from the live Internet, but there are two things we do know: 1) The responses don't always come from the model. There seem to be multiple guardrail agents specialising in different things. If you can have a guardrail agent, you can have an agent that detects a riddle with a known stored answer and bypass the model. I'm speculating, of course. Just a conspiracy hypothesis 😅 2) There's a large number of humans whose job it is to manually train the system via reinforcement learning.
2LegHumanist
2024-03-02 09:42:40 +0000 UTCI went and watched the vaush break down video. With my pure little eyes, and my pristine little ears.
Ymirsdreams
2024-03-02 09:11:38 +0000 UTCI am one of those people who needs background noise to fall asleep. So I have fallen asleep while listening to DtG but it was not ideal because I woke up a few times to a very vivid nightmare that Chris was telling me off! I've now swapped my falling asleep podcast to Misquoting Jesus with Bart Erham and Megan Lewis who both have lovely calm voices!
Emma
2024-03-02 08:55:49 +0000 UTCThat's why I avoid testing it with published puzzles, but rather dream up 'weird' ones - things that are not the kind of thing that people write about or think about much, but just know. It's a lot more hassle than grabbing a test-bank, but necessary. Many of the larger tests being administered are declarative knowledge about academic subjects that often appear in textbooks. Much better IMO are things that are never written about - like, IDK, "How easy would it be to make a snowman out of tomatoes?". I don't think it's an online learner. Yes, the trained weights get updated from time to time, but I'm pretty sure it's not adapting to the live internet. Note the "my training data is current up to XXX" disclaimers. Yep, I'm aware that reinforcement learning is incorporated. All that said, OpenAI is not open, and we can't be 100% sure exactly what's going on under the hood.
Guruspod 2
2024-03-02 07:40:59 +0000 UTCRegarding the riddles ChatGPT can and can't answer, you have to remember it's more than just an LLM. There are all sorts of layers and guardrails and multiple LLMs. There's also an army of humans influencing its training manually through human-in-the-loop reinforcement learning. There have been countless suspicious examples of riddles like those mentioned, that stump chatGPT one day and suddenly not any more once the riddle has gone viral. There have also been cases where a riddle ChatGPT became suddenly capable of answering after being published on twitter, was updated with different particulars, keeping the same logic. That's all it took for ChatGPT to get it wrong. It also turns out that if you want to find a programming puzzle that chatGPT cant answer, you just have to find one with a solution that was never shared online.
2LegHumanist
2024-03-02 07:16:00 +0000 UTCThe most useful insights and skepticism of AGI claims, and the spurious claims made about LLMs, come from cognitive scientists. Especially for a black-box system like an ANN. They really know how to probe them. They are, afterall, experts on cognition. I'd go as far as to say ML researchers are speaking outside of their field when talking about how close we are to AGI or whether this or that technology is a stepping stone. Regarding the other definition of AGI, aren't we just talking about extrapolation? If so, I think it's better framed as the extrapolation vs interpolation problem.
2LegHumanist
2024-03-02 06:24:20 +0000 UTCI don't think Mearsheimer thinks the "realist" school of politics espouses amorality. It is descriptive of how nations behave, that's different from approving such behavior, very different. I really think Matt got that point wrong. Or at least how I took what Matt said. I liked the point on Glenn Loury. Kind of glad he was humbled, but still appreciate his humility. He should be a little more humbled, I saw a leftist bio of George Floyd and it didn't make him into a hero whatsoever, the leftist take is more that the system failed him, it got him to aspire to be a basketball star for which he didn't have the athletic talent, then provided no other path to solid employment. Am just halfway through. Still trying to work out the strategic disclaimers and honest ones.
Kat
2024-03-02 06:15:45 +0000 UTCThe Patreon is a kind of safe haven away from all the chuds. And I forget sometimes what's lurking out there. If I can't stop myself from examining the youtube comments I may need to lock myself away in a one of Rogan's deprivation tanks for a bit.
JustTheWorstEver
2024-03-02 02:23:00 +0000 UTCAppropos of nothing I just want to comment that LinkedIn just tried to get me to follow both Brene Brown and Andy Hubermann in the same spam email. We're being assaulted by grifting gurus everywhere we go because they all possess some magical cloak of reputability on social media. It's poignant after seeing all the grouchy comments on your new youtube video defending JP and Bret. They of course being the forces for light and reason while you two are nothing but petty losers who've accomplished nothing, making your vicious ad hominem attacks, "guilty of everything you accuse these men of being," lol.
JustTheWorstEver
2024-03-02 02:21:37 +0000 UTCi became Vaush’s fan when I argued with someone in like 2020 how he’s probably a pedophile and took time to watch source video to write better comment; video was pretty based and clip made sense in context
aneladgam_varelse
2024-03-01 23:55:10 +0000 UTCLooking forward to this one! I gave up on his show long ago, too much condescension for my taste. And I can't take anyone who is so convinced of the many worlds interpretation seriously.
Jason Etheridge
2024-03-01 23:43:18 +0000 UTCOh I had adult braces to! Mine also weren't 100% successful, but if your teeth are crooked it's not noticeable at all.
Emma
2024-03-01 23:35:11 +0000 UTCI look forward to some truly deranged Weinsteinian pontification & conspiracism from red-pilled menace Sean Carroll.
Rasterisk
2024-03-01 23:15:52 +0000 UTC
