Everything Was Already AI (Rough Cut)
Added 2025-11-17 21:54:34 +0000 UTCFeedback welcome, hope you enjoy this video which was a lot of fun to make!
References (in rough order of appearance)
How to Make Realistic Predictions About AI, Tantham https://curveshift.net/p/how-to-make-realistic-predictions
https://www.youtube.com/watch?v=8enXRDlWguU&
‘Large AI models are cultural and social technologies’, Farrell et al.
Artificial Intelligences, Herbert Simon
Debunking Economics, Keen
https://www.mic.com/articles/107896/scientists-finally-prove-why-pop-music-all-sounds-the-same
The Dorito Effect, Shatzker
https://www.youtube.com/watch?v=lRq0pESKJgg&
The Stock Market is a Conventional Wisdom Processor: Why Trump’s Tariffs Crashed the Stock Market While the Trump Musk Payments Crisis Hasn’t (Yet), Tankus https://www.crisesnotes.com/content/files/2025/04/The-Stock-Market-is-a-Conventional-Wisdom-Processor-Why-Trump-s-Tariffs-Crashed-the-Stock-Market-While-the-Trump-Musk-Payments-Crisis-Hasn-t--Yet-.pdf
https://www.youtube.com/watch?v=Gqlbn2nPO-A&
https://www.theatlantic.com/ideas/archive/2025/09/job-market-hell/684133/
https://www.youtube.com/watch?v=gJW4-cOZt8A
https://www.youtube.com/watch?v=GW2Zr8Q6Xqw
https://www.youtube.com/watch?v=0xeMak4RqJA
https://www.youtube.com/watch?v=_zfN9wnPvU0
https://www.youtube.com/watch?v=9E6p3J9dko8&
An Existing, Ecologically-Successful Genus Of Collectively Intelligent Artificial Creatures, Kuipers
AI Integration Is the New Moat, Tim O’Reilly https://www.oreilly.com/radar/integration-is-the-new-moat/
https://www.youtube.com/watch?v=qvpw4_O25eU
https://www.youtube.com/watch?v=y3HpdNGvJDc&
notes on the industrialisation of decision making, Davies https://substack.com/home/post/p-177016285
the only message the channel can carry is a scream, Davies https://substack.com/home/post/p-151264334
The AI Circular Economy, Blakeley https://substack.com/@graceblakeley/p-176721998
The Case Against Generative AI, Zitron https://www.wheresyoured.at/the-case-against-generative-ai/
The Map is Eating the Territory: The Political Economy of AI, Farrell https://www.programmablemutter.com/p/the-political-economy-of-ai
https://www.youtube.com/watch?v=d8reiauyQCM
Further reading
https://open.spotify.com/episode/4pWuwQq8M8Gzf9F9U0AYZW
https://www.youtube.com/watch?v=wjZofJX0v4M&list=WL
https://www.youtube.com/watch?v=6pzLhWCxH_g&list=WL
AI As a Normal Technology, Arvind Narayanan & Sayash Kapoor https://knightcolumbia.org/content/ai-as-normal-technology
Comments
Great video, Cue the brain spawns from Futurama at the end! The bit were they finalize learning everything and just end the universe. On a more serious note, accountability for leaders throughout history was much more real..
pierre-luc gelineau
2025-12-03 23:09:38 +0000 UTCThanks for these critiques, this makes sense although I do still think that the main point is correct. There is another study from around the same time - Serra et al (2012) - which uses pitch, timbre, and loudness, and finds that music is often quite similar over time but has homogenised somewhat over time too.
Unlearning Economics
2025-11-26 12:32:44 +0000 UTCThat does sound extremely interesting, yes I do get the explore and exploit example. I wonder to what extent human behaviour tracks algorithms more when we have strong cultural technologies that steer us towards certain goals...
Unlearning Economics
2025-11-26 12:29:50 +0000 UTCAh yes!
Unlearning Economics
2025-11-25 12:30:44 +0000 UTCIt just came to my mind how I the Simpsons analogy can be used for children's animation too. Whether it's infinite spongebob or 3D slop with empty backgrounds, it looks like no one is willing to ask for quality in animation. Last wave of creativity happened more than a decade ago, no one cares what children are watching, the focus is on feeding them more slop, faster. Original 2D animation is a form of art that we should protect, whether you can sell merch for it or not. It's also a matter of cultivating taste, a desire for aesthetic beauty or ugliness (like the 90's and early 2000's raunchy styles) which is shamefully no longer sought after. Everything looks like everything else. etc. etc.
Martian Mochi
2025-11-23 22:13:50 +0000 UTCA masterpiece packed with some of the best humor and incorporating so many video essays made on this topic, youtube feels like a proper college syllabus now 😊! I don't know if anything can be removed from this 'rough cut', the video has great flow as is.
Martian Mochi
2025-11-23 22:04:01 +0000 UTC!!!!!! I don't know if you've across this book yet or not, but something you said around 15 minutes in about the comparison between the way LLMs break language down to stat models and the way that humans transmit information to each other and then use culture to make sense of it reminded me of a cool book I read several years ago called "Algorithms To Live By" by Brian Christian and Tom Griffiths. I think Tom is a computer science professor somewhere and Brian Christian does experimental psychology. I'll start by saying there's some really stupid dating advice in the book, so ignore that. But the really cool idea in the book is that they note the ways that machine learning often mimics the way humans tend to act in the agrigate and that individuals often act as though we are following relatively simple computational rules. Like they have this example about the pile of clothes by your bed being a kind of "cache" of your wardrobe because it's the things you accessed the most recently and, therefore, the things you're most likely to want to access again in the near future. They also talk about the ways that things like Hollywood sequals are very similar to explore-exploit algorithms where if you want to find the optimal outcome of something you spend the first x amount of time exploring the available solution space and then spend the remaining time "exploiting" what you found in the exploration. So, if Hollywood found something that works in their "explore" phase, it makes more sense for them to exploit it with sequals during the later phases of trying to make money than to spend time searching again. There's also some examples of how human behavior in aggregate can give you an indication of if you should be on guard for financial or political crises but that it can be hard here because while humans in aggregate often act like they are using fitting models to make group decisions, humans and computers are equally prone to over fitting when they have access to too much information and no clear way to sort through it to decide which information is important, which is where you get people seeing patterns that don't exist or something like that (personally I think this is where conspiracy theories come from). Anywho, I thought you might like this book. It's really information dense and you'll probably find that most of the things in there are known examples, but they put it together in a way that I think is novel and the synthesis between computer science, mathematics, and psychology is super interesting. I know that some of those same ideas are influential in areas like relationship science, too, specifically game theory.
Kodak
2025-11-23 19:19:42 +0000 UTCWhat a great video! Proud to be a patron of yours :) A small correction, however: at 1:34:45 – you probably mean Copilot here, not GitHub?
Vladimir Sechkarev
2025-11-23 18:11:02 +0000 UTCOh, one side point: Simon worked on a _very_ different kind of AI than is common today. He did automated mathematical proofs, as far as I know. No data and no machine learning involved - just a lot of formal logic with strict guarantees.
DrAmi
2025-11-22 19:09:36 +0000 UTCGotta read more Herbert Simon, it seems. :) I think it was a little hard to follow during the longer read sections from Simon's work, but that may just be my declining attention span and fried brain. Very much worth listening to all of it, though.
DrAmi
2025-11-22 19:06:56 +0000 UTCThis was really good! I love it when a video goes beyond analyzing problems to actually suggest workable solutions.
Alice
2025-11-21 19:58:09 +0000 UTCi would love to see a video on all the "we have made strides towards solving X" they soikund like high level nigh intractable problems.
Kitrana
2025-11-20 22:01:25 +0000 UTCYou know, I did think of this but animators are just expensive. Maybe I'll ask my editor what he can manage.
Unlearning Economics
2025-11-20 16:24:42 +0000 UTCdo they appear in this vid!?
Unlearning Economics
2025-11-20 16:23:42 +0000 UTCThat was the exact purpose of the vid!
Unlearning Economics
2025-11-20 16:23:37 +0000 UTCHave to agree, it was a real banger. I liked the perspective and also the analogies throughout. Especially the focus on accountability and trying to apply cybernetics as a possible solution felt very resonant to me. I also enjoyed the timing and amount of snarkiness towards Eyepatch Wolf and even about the reaction to Alexander Avila's video about AI. Thanks for the nice video!
ValentineSmith
2025-11-20 12:52:52 +0000 UTCThis is my favorite video yet. Seriously, this is good stuff. And I hope this gets reach because we all feel things being wrong but don't have words or solutions
Pixie Plays
2025-11-19 22:13:34 +0000 UTCLoved Vasnetsov’s Bogatyrs on the wall
alina khalitova
2025-11-18 22:26:25 +0000 UTCbanger
LIFEBOD
2025-11-18 20:01:36 +0000 UTCThat music study sounds weird. Using the quantity and commonality of musical instruments to assess the "accessibility" of music is strange, no? Because 1) you can use a piano to play Chopsticks or you can use it to play Chopin - the instrument itself will have almost nothing to do with how complex or how popular the music is, presumably; 2) a lot, if not most, of pop music these days is made with just synths, which can be used by a single musician to achieve a wide variety of sounds; 3) the commonality of music instruments also affects how many musicians there are playing them: there aren't that many rock songs with a bagpipes solo in them (shout out to Under the Milky Way for being the absolute banger it is) because there aren't that many wannabe rock stars learning the bagpipes to start a band.
Boris Vorobev
2025-11-18 17:12:27 +0000 UTCEmboldened by past experiments, UE is consolidating his position of power and abusing it to advance a one-sided, pro-simpsons agenda on his helpless viewers
SynLynx
2025-11-18 08:59:10 +0000 UTCI really enjoyed the video, thanks a tonne for your continued bangers. 1. (3:17) The term "AI" definitely isn't restricted to language modelling so I wouldn't write "large *language* models" 2. (7:53) I wanted to comment for this part, I disagree strongly with the framing of LLM training as like a genetic process. The vast majority of AI training happens under mini-batch gradient descent which is a guided local search. 3.1. (6:49) This part is earlier but I think "brute-force" is not a good term here, as a brute-force solution (in my mind) describes a process where model parameters/formations are sampled exhaustively rather than updated/guided according to the training process. 3.2. Later on in this section, the "half of the sentence is known" and "half is unknown" description of training is a fine example for the video but I don't feel the first-half/second-half setup accurately describes any particular training regime (as opposed to hiding random subwords of the sentence, or hiding every next-subword in a sentence and predicting simultaneously)
Green Apple Pie Sucks
2025-11-18 05:21:43 +0000 UTCThe local optimization hill analogy could use a proper visualization. Though maybe it deviates from your general style with the Simpsons and all too much
Väinö Yrjänäinen
2025-11-17 22:33:10 +0000 UTC
