December 2021 News
December 2021 Gwern.net newsletter with links on TODO
December 2021’s Gwern.net newsletter is now out; previous, November 2021 (archives). This is a collation of links and summary of major changes, overlapping with my Changelog; brought to you by my donors on Patreon.
Writings
-
Gwern.net: GPT-3-powered (paper; API) “similar links” feature added to popups/indexes
Links
AI
-
“Learning robust perceptive locomotion for quadrupedal robots in the wild”, et al 2022; “Agile Locomotion via Model-free Learning”, et al 2022; “Concurrent Training of a Control Policy and a State Estimator for Dynamic and Robust Legged Locomotion”, et al 2022; “Robot peels banana with goal-conditioned dual-action deep imitation learning”, et al 2022 ( video)
-
“Automated Reinforcement Learning (AutoRL): A Survey and Open Problems”, Parker-et al 2022
-
“M5 accuracy competition: Results, findings, and conclusions”, et al 2022 (machine learning forecasting methods have finally convincingly beaten statistics in an M-competition)
-
“StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasets”, et al 2022 ( StyleGAN3 for 1024px-sized ImageNet synthesis, using progressive growing+projected GAN+ViT layers)
-
“DishBrain: In vitro neurons learn and exhibit sentience when embodied in a simulated game-world”, et al 2021 (amazing feat, even if it doesn’t learn from images & it’s much slower & worse than DRL)
-
“Word Golf”, Eric Xia (traverse the GloVe graph of word embeddings from a starting point to a target word)
-
“ERNIE 3.0 Titan: Exploring Larger-scale Knowledge Enhanced Pre-training for Language Understanding and Generation”, et al 2021 (Baidu scales up ERNIE 3.0 to a 260b zh Transformer-XL + adversarial loss + knowledge graph + distillation + CTRL-style control; still training, on 1920104ya NPUs, but sets many zh benchmark SOTAs, beating Yuan 1.0)
-
“WebGPT: Browser-assisted question-answering with human feedback”, et al 2021 ( blog; OpenAI Universe was a few years too early—but nevertheless, “tool AIs want to be agent AIs”)
-
“Harmful content can evolve quickly. Our new AI system adapts to tackle it”, FB (large multilingual meta-learning RL-tuned Transformer for rapid few-shot censorship of posts); “Few-shot Instruction Prompts for Pretrained Language Models to Detect Social Biases”, et al 2021 ( Megatron-NLG-530b)
-
“Retrieval-Enhanced Transformer (RETRO): Improving language models by retrieving from trillions of tokens”, et al 2021
-
“A Mathematical Framework for Transformer Circuits”, et al 2021/ “In-context Learning and Induction Heads”, et al 2022 (invisible phase transition in Transformer internal algorithms after enough training)
-
Revisiting “The Brain as a Universal Learning Machine”; “Brain Efficiency: Much More than You Wanted to Know”, Jacob Cannell
-
“AI Chains: Transparent and Controllable Human-AI Interaction by Chaining Large Language Model Prompts”, et al 2021 ( GPT-3/LaMDA)
-
“Recent Advances in Natural Language Processing via Large Pre-Trained Language Models: A Survey”, et al 2021
-
“Mining for strong gravitational lenses with self-supervised learning”, et al 2021
-
LG’s Exaone (L-Verse architecture): a 300b-parameter multimodal Transformer trained on 600b Korean-English tokens (?) & 250m images
-
“LaMDA: Language Models for Dialog Applications”, et al 2022 (scaling, retrieval, finetuning); “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models”, et al 2022 ( blessings of scale: LaMDA inner monologue capabilities abruptly emerge at ≥100b-parameters)
-
“Vision Models Are More Robust And Fair When Pretrained On Uncurated Images Without Supervision”, et al 2022 (scaling SWaV to 10b-parameter RegNet CNN w/n = 1b Instagram images)
-
Announcing GPT-NeoX-20B: largest dense public Transformer model
-
“Moore’s Law, AI, and the pace of progress”, Veedrac (hope for hardware progress)
-
minDALL-E, Kakao Brain (1.3b text-to-image generation model trained on 14 million image-text pairs)
Genetics
Everything Is Heritable:
-
“Genome-wide association meta-analysis identifies 29 new acne susceptibility loci”, et al 2022; “Every Pore on Your Face Is a Walled Garden: A close examination of human skin found that each pore had a single clonal variety of bacteria living inside”
Recent Evolution:
Engineering:
Statistics/Meta-Science
-
“This scientist accused the supplement industry of fraud. Now, his own work is under fire” (Gelman; revisiting that 201311ya supplements paper); “Star botanist likely made up data about nutritional supplements, new probe finds: After 2022 exoneration, fresh inquiry concludes Steven Newmaster fabricated data that questioned quality of echinacea, ginkgo, and other substances”
Politics/Religion
-
October Crisis (200+ bombs, 23 terrorists already in prison for crimes including at least 7 deaths, bank robberies, stealing tons of dynamite, at least 2 failed attempts to kidnap diplomats, murdering the hostage provincial deputy premier, & escaping to Cuba in order to foment a revolution with bonus KGB involvement; result: “By 198242ya all the convicted participants had been paroled & all of those sent to Cuba had returned to Canada, some completing short sentences”. The 1970s—you had to be there.)
-
“Diversity Training Goals, Limitations, and Promise: A Review of the Multidisciplinary Literature”, 2022
Psychology/Biology
-
“The evolution of quantitative sensitivity”, et al 2021 ( media; the human brain isn’t special in counting—just big)
-
“Like a Lemon to a Lime, a Lime to a Lemon” (what even is a ‘lemon’? We just don’t know. The other day I accidentally bought Cara Cara oranges which are delicious; where did it come from? It just mutated off a tree in Venezuela… maybe.)
-
“Results of a 5-Year n-of-1 Growth Hormone Releasing Hormone Gene Therapy Experiment”, Hanley et al 2021
-
“The lungfish cocoon is a living tissue with antimicrobial functions”, et al 2021
-
“Polytetrafluoroethylene (PTFE) Ingestion as a Way to Increase Food Volume and Hence Satiety Without Increasing Calorie Content”, et al 2016 (for further mad food science: nicotine toothpaste, oil → margarine, perfume, bog butter, what is the oldest possible food)
Technology
-
“A Personal View of Average-Case Complexity”, Impagliazzo 199529ya (the 5 Worlds of computational complexity: Algorithmica, Heuristica, Pessiland, Minicrypt, & Cryptomania)
-
“Understanding Sources of Inefficiency in General-Purpose Chips”, Hameed et al 201014ya (where do ASIC speedups come from?)
-
“From fish out of water to new insights on navigation mechanisms in animals”, et al 2022 ( video)
-
“Surprisingly Robust In-Hand Manipulation: An Empirical Study”, et al 2022 (hand-designed primitives for inflatable hand: learning-free, open loop, but still reliably manipulate cubes)
-
“A Large-Scale Characterization of How Readers Browse Wikipedia”, et al 2021 (WP readers visit 1.5 pages, almost all by following cross-page links; do we underestimate the value of a hyperlink?)
Economics
Philosophy
Fiction
Miscellaneous
-
“Iron and the Soul”, Henry Rollins 199430ya: “Try to lift what you’re not prepared to and the Iron will teach you a little lesson in restraint and self-control.” (“Black Flag’s change in musical style and appearance alienated many of their original fans, who focused their displeasure on Rollins by punching him in the mouth, stabbing him with pens, or scratching him with their nails, among other things. He often fought back, dragging audience members on stage and assaulting them…In a 200123ya interview with Howard Stern, Rollins was asked about rumors that he kept Cole’s brain in his house.”)
Books
Nonfiction:
Fiction:
Film/TV
Live-action:
-
The Third Man (supposedly one of the greatest films of all time, and something of a disappointment. A long slog with a sap protagonist who I quickly started rooting for being sapped in a back alley, famous tilted shots less interesting than The Cabinet of Dr. Caligari 29 years before, a plot driven by one of the less convincing sociopath characters of film whose most famous line about cuckoo clocks turns out to be delivered in a slapdash and hurried manner and whose central Ferris wheel scene is claustrophobic & bland, a climactic sewer chase scene notable mainly for being both repetitive & confused—I doubt even the most careful viewer could draw out a map of the maze… A few scenes undeniably work, like Harry Lime grasping through a sewer grate as he dies or the final scene where the protagonist is snubbed, but I’m hardpressed to otherwise explain why anyone would elevate this above other film noir or movies like M.)
Animated:
-
I was shocked to note the 10th anniversary of Puella Magi Madoka Magica recently. I had watched it during the original broadcast when, after episode 3, the buzz began building that (weird faces or no) Gen Urobuchi had done it again, and picked it up after everyone went crazy with the time-loop reveal to watch the ending, which more than lived up to expectations (did it really air on Good Friday, and get delayed by an earthquake‽). It was a series to rewatch someday, and since I hadn’t rewatched it or thought much about it since watching Madoka: Rebellion back in 20141 & reading the military SF fanfiction “To The Stars” (still updating as of 2021, still nowhere near an ending), this made it a good test-case for my media forgetting curve notes: at 10 years, I should have forgotten most of it and be able to watch it almost for the first time.
Madoka turned out to be even better than I remembered. I recall the overall plot reasonably well, but all the details were fresh and surprising, and with the ending in mind, I could properly appreciate all of the foreshadowing and details. The soundtrack remains one of the best anime soundtracks ever made, and the tracks work perfectly in context. The famous witch animations have lost none of their impact, and it’s amazing just how tightly written and well-constructed Madoka is. It does far more in 12 episodes than bullshit like Legend of the Galactic Heroes can do in 110 episodes.
Every moment is laden with import. For example, it heavily flags that it’s going to be a subversion/deconstruction experiment by twisting almost every standard element of the highschool romcom/magical girl—Madoka wakes her mom up instead of being woken up by her; her dad is a househusband and her mom works, instead of the usual dad or two-working parent household; Madoka discusses boys with her mom & looks nice for her secret admirer at school, who will turn out to be a girl (Homura). Kyuubey idly speculates later about wishes being an energy source so powerful that it can break the laws of physics & how Madoka has accumulated power equivalent to a god, and in episode 3, Madoka talks about how she has no wish, because just being a magical girl and helping everyone would be her wish—which is how the series ends, with her turning into a boddhisattva. Subtly depicted—blink and you’ll miss it like I did the first viewing—but critical is the fact that the timeloop adds weight because this is the last loop before Homura succumbs to despair as no matter how she speedruns, she can’t defeat the greatest witch, and that is why we are seeing this specific timeloop, because it is not simply the ending, but the last one, played for keeps. (Homura’s gradual degradation and loss of humanity, while accidentally simultaneously empowering Madoka every time to the point where her inevitable turn into a witch dooms humanity, is well-portrayed in the loops, and reminds me of review_The Empty Box and The Zeroth Maria_’s point that being awake in a timeloop is as big a vulnerability as it is opportunity.) Mami’s death is heavily foreshadowed by her emotional development in trying to be a kind senpai giving her kohais the start he never had—heavily remiscent of Asuka in 2.0, complete with white-mask-faced Charlotte feeding on the corpse like the MP Evas in End of Evangelion…—blatant when you know it’s coming but I was still shocked. (“…Nobody wants to die that way. People die of disease and accident. Death comes suddenly”.)
Nor is that the only instance of EoE influence; like every sekaikei anime, EoE’s influence runs deep (glowing figures tanging the entire population of magical girls, check; cosmic figure of giant witch embracing the earth, looking exactly like Second Impact’s wings or the Giant Naked Rei, check; naked Madoka floating in orange ocean which is definitely not the Sea Of LCL while strange alien entities existentially interrogate her, check; imaginary world where no one knows Madoka, check; ‘the end you wished for’, check), but where most are crushed by it and unable to do anything but retread its plot beats to restore a normal post-apocalypse (weakened for being so second-hand and so much less committed to existentialism than EoE’s notoriously harsh ending), Madoka is able to both transcend it and produce an emotional impact that will stay with the viewer.
Someone, somewhere, is fighting for you.
But the faces still look weird.
<– 13:53 < gwern> …openly aligned with the military, thanks to a cataclysmic alien invasion that forced magical girls to reveal their existence to the world.’ https://www.animenewsnetwork.com/feature/2021-07-07/madoka-magica-isnt-the-same-without-gen-urobuchi/.172767 senpai noticed ; 17:31 < gwern> https://www.youtube.com/watch?v=6lcottCoAjk “I should have wished for a friend with balls” 11:22 < gwern> so much foreshadowing in Madoka: https://imgur.com/MJ6LP I feel like an idiot
12:22 < gwern> ‘I have a friend who was once unemployed for a long time and didn’t have any motivation to find a job. Recently he found his job and happy with it. I’m utterly surprised and ask what motivated him. “So I can buy Madoka BDs without hurting my balance” is his response. True story.’ ‘truly, she works in mysterious ways’
12:38 <@gwern> http://paptorts.livejournal.com/29337.html recipe ’…Procedure: 1. Bone the pork shoulder. You can go about this several ways, but I usually start by complimenting its dress and 21:30 <@gwern> haha so I was googling the madoka phrase and somehow I wound up… here: http://danbooru.donmai.us/pool/show/2862
13:03 < Tuxedage> http://wiki.puella-magi.net/Mathematics_of_Madoka_Magica 19:25 < klfwip> Yeesh, some people take this really seriously. http://listlessink.wordpress.com/201113ya/04/23/birth-of-a-goddess-madokas-path-to-nirvana-a-study-of-buddhism/
12:33 <@gwern> … Shakespeare than W. W. Jacobs.’ https://steepholm.livejournal.com/401858.html 02:41 < feepbot> feep: https://tvtropes.org/pmwiki/pmwiki.php/FanficRecs/PuellaMagiMadokaMagica 04:21 < feepbot> gwern: https://www.youtube.com/watch?v=d-FcS0EEnYY Madoka Magica In 30 Seconds - YouTube 07:07 < two2theheadPC0> Should be here: https://wiki.puella-magi.net/Madoka_Magica_Soundtrack
05:24 < skybox> Q: Madoka seems to be inspired by Goethe’s Faust and in Psycho-Pass you quoted works from Max Weber and other German writers. How did you discover them and how did they inspire you? 05:24 < skybox> A: The designers from SHAFT thought of inserting the quotes from Faust. I only noticed them when people approached me asking about them. My main inspirations are eroge and classical literature. –>
-
A spiritual sequel to The Tatami Galaxy also directed by the inimitable Masaaki Yuasa; if you ever wondered what Tatami Galaxy would be like if not confined to single episodes which had to reset & if all the ingenuity was concentrated into a single intricate picaresque plot with no adult supervision—Night is the answer.
Bursting with energy and an intertwined plot, I had to watch it twice to properly appreciate it. (To give an example: at the climax of the second act, the guerrilla theater group being chased by the student government’s secret police sets up the grand finale of their play which is intended to reunite the dramaturge with a girl he fell in love with when apples fell out of the sky and hit them on the head simultaneously; at the climax while he sings out to her, the girl finally appears—and arrests him, as she was actually head of the secret police head who enjoyed crossdressing—being resolved by koi fish falling out of the sky to hit him on the head again, and much later we realize that both apples & fish falls were caused by the lonely misanthropic antagonist, who believed he was alone & didn’t matter to anyone, but has all along been the cause of the sub-plots.)
Night is now my favorite Yuasa work, and one of the best anime I’ve watched in a while. It’s tragic that it appears to’ve passed with such little notice (was it the unmemorable title? or did people think you had to watch Tatami Galaxy? You don’t, even if it helps you a bit to slot in characters given how fast things move).
-
Devilman Crybaby (2018; 10 episodes; Netflix)
Interesting failure. Entirely by accident, forgetting that he directed it, I watched this next. Night Is Short was a Masaaki Yuasa work critically applauded but watched by no one; Devilman Crybaby (DC) turns out to be the opposite. But DC is rarely boring. Whatever else one says of it, one must grant it that.
Plot/theme summary. To recap DC: Devilman was a famously gory and raunchy manga by Go Nagai; it influenced Hideaki Anno & Evangelion, particularly End of Evangelion. DC follows the ordinary boy Akira, of no talent, whose defining trait is a sentimental heart; a childhood friend, the unnaturally blond Ryo, suddenly returns. Ryo was a strange child, who Akira found abandoned and took pity on, and seemed emotionally crippled—unable to try to save dying kittens & instead urging its immediate euthanasia because it was ‘weak’—but Ryo was so gifted, he was studying overseas. (You can tell it was written pre-2000s just from that trope, incidentally, given Japanese involution.) Ryo confides in him the discovery that the world is being controlled by a cabal of “devils”, supernatural monsters who take over human bodies, and hedonistically kill and eat humans. Investigating a rave where devils possess hosts, Akira is himself possessed by a devil, but his heart is too kind to become a monster, leaving him in control of the powers (he dubs such a failed possession a “devilman”). Akira encounters several devils, all of which fight him, and refuse to surrender or renounce violence, forcing him to kill them in self-defense (of course). Those devils apparently have extensive backstories, but DC is in such a hurry (not even 1-cours!) I’m not sure we get their names, much less care about them as antagonists; there are also some minor bits with gay characters which come off as tacked on, and a number of rapping characters who are inessential but interesting. Finally, Ryo concludes that the devils are too entrenched to continue fighting covertly, one by one: if humanity is to survive the devils, there is no choice but to end the masquerade and trigger a global war. He ruthlessly engineers a massacre by the devils at a major sporting event on live TV, triggering a total war of extermination. The devils fight back, now determined to end the threat of humanity, subverting and dividing humanity. Anyone could be a devil. Civilization begins to collapse in internecine war, and indiscriminate attacks and executions. Akira is unable to reason with anyone, and is persecuted by the masses, and his family butchered. (Oddly, the family cat turns out to be a ‘devilman’ too. This is never explained.) Partway, the devils reveal the truth to Ryo: ‘Ryo’ is not a human at all, but a fallen angel, the hermaphroditic Satan, expelled from Heaven by the Kannon-esque Buddhas (the cosmology can’t seem to make up its mind if it’s Buddhist or Christian) for his moral failings; he fell to Earth, where the devils had evolved all their powers, “red in tooth and claw”, by survival of the fittest; he continued his rebellion in pitting devils against each other to find the strongest, so the Buddhas/angels destroyed the world, killing them all; but now, eons later, they have been able to reincarnate thanks to humans, and Satan chose the strongest of the devils to empower his beloved childhood friend to oversee another war to reveal the strongest. With the nigh-invincible Satan on their side, the devils fight humanity to a standstill, and Akira & the devilmen join in the fight. But it is futile and only accomplishes a mutual slaying. In the infamous ending scene, Satan lays next to Akira on barren sea rocks, the only survivors, and as Satan talks to Akira and the camera pans out, he realizes that Akira is only the upper half of his body (which was all that was visible in-frame) and has died—because Satan killed him. Satan breaks down weeping as he suddenly understands caring about someone else and the pain of loss. Far above, the Buddha/angels send down destruction, presumably killing Satan and starting another cycle of regrowth of life & reincarnation, but with, perhaps, some slight redemption of Satan, and some hope that the next cycle will be better.
NGE. The connection to Evangelion is immediately clear (the appearance of Evangelions, monster-of-the-week leading into apocalypse, man-is-the-real-monster, fusing with monsters to harness their power, ancient alien civilizations returning, cycles, Christian-Buddhist syncretism), particularly if you read a description of the scrapped EoE ending called “Last B”. Devilman, EoE, Last B, and Concurrency all are orbiting around a central point about the grimness of reality, but for all the horror, a ray of hope in progress. In Devilman, the apparently endless suffering and futility is allayed by the hope that one day, the Boddhisatvas will save all sentient beings, and that even Satan will be redeemed one day. So, quite interesting for me to watch. Like watching Space Battleship Yamato, I felt I should’ve watched or read a Devilman long before.
Right story, wrong director. But that is not to say it is good. I would say that DC is a failure at its goal of making one cry, even if Satan does in the end. The fatal flaw is not hard to see: it’s just that Yuasa is one of the best anime directors in the world, and one of the worst Devilman anime directors in the world. (To be a fly on the wall at Netflix when this was decided… Flunkie: “I feel I should ask—has anyone here ever watched any of Mr. Yuasa’s anime?” Executive: “He’s a talented guy. It’ll be fine. And do you know how hard it is to book anyone these days?”)
Comedy ≠ Tragedy. Comedy and tragedy may be written with the same letters, as Democritus put it, but they are not drawn with the same lines. The hyperkinetic Yuasa animation style is perfect for optimistic bildungsroman or picaresque comedies or sports drama like Tatami Galaxy or The Night Is Young or Ping Pong, where any darkness is quickly dispelled and the loose animation conveys movement and hyperbole. DC, on the other hand, is fundamentally about a pessimistic plunge into the darkness of existence, the cruelty of sentient beings to each other, and the apocalypse, relieved by a single ray of hope. Plus cool monsters fighting with superpowers.
Bathos, not pathos. DC drew criticism for all the titillating violence and sex (I have never seen an anime quite so enthusiastic about female masturbation), but what should’ve drawn criticism is how un-titillating any of it is. The sex is a ten-year-old’s imagining of it, thoroughly repulsive, and is more likely to make one want to swear a vow of celibacy than threaten one’s celibacy. Yuasa’s touch continually twists tragedy into comedy and bathos, and the lightness of movement undermines any impact: fights lack body, coming off as just a lot of wiggly lines bouncing in mid-air, and lacking interest. (What are Akira’s powers? Mostly just tearing stuff in half, or punching them. The ‘king of devils’ sure is boring.) Suspension of disbelief fails, and scenes like Akira running through a rave randomly stabbing people with a bottle to summon devils, which ought to be shocking, come off as a shrug. In another scene, the Christian father character (not played for laughs or criticism but sincerely—highly unusual) discovers his son has turned into a monster who has killed & devoured his mother, and, now a broken man, raises his gun to shoot, only to lower it in agony; deeply moving the first time! Also the, er, second time. Not so much the third. By the fourth or fifth, all pity has fled and I was involuntarily laughing. When the heroine runs down the side of a river, hunted down by humans (who have become the real monsters) in trucks who shoot her, one can’t help but think—“have you tried running in not a straight line? That would probably work better.” (Since this was a Netflix production, it comes with extensive dubs; I briefly tried the English dub and the voice actors immediately came off as not taking it seriously and hamming it up, as so often with non-Japanese dubs, which would’ve made all this worse.)
Thus, rarely boring. But good? Eh.
-
Odd Taxi (one of the standouts of the past season: an ordinary taxidriver in contemporary Tokyo who happens to live in a world of talking animals gets caught up in a Durarara!-esque web of social-media-enabled plots and must navigate his way out, although to everyoe around him, the biggest mystery is him… A solid character-based drama which makes effective use of the animal gimmick)
What distinguishes an ‘homage’ from an ‘imitation’, or worse, a ‘plagiarism’/‘ripoff’? Why does the extent to which the ending of Flip-Flappers copies End of Evangelion leave a bad taste in my mouth, while Kill la Kill’s similar use is neutral, and EoE’s own use of Battle of Okinawa (perhaps via Space Battleship Yamato) feels positive?
Deception and credit can be a part of it: Flip-Flappers feels like it’s hoping you won’t notice, while KlK expects you to know, and that does make a difference; but on the other hand, it’s not the only difference, because EoE can hardly expect you to have even heard of some 197153ya war film (and still likewise for Yamato because while it was popular, it would be unrealistic of Yamato’s makers to expect kids to have seen an adult movie from 4 years previously—they could hardly stream it on Netflix). When Simpsons makes a Rashomon reference, it surely flies over the head of >95% of viewers (I know it did for me until I watched a bunch of Kurosawa movies for school and then caught a rerun) or re-enacts a movie plot without any explicit signaling of which movie, this is all par for the course and indeed a good part of why one watches it. ‘Being lazy’ may be part of it, but also appears inadequate: often, copying something may be quite a lot of work! EoE is hardly an easy act to follow; there must be countless alternative endings which would just be way easier to animate. (Writing is hard, but it’s not that hard.) Lack of novelty or new contributions is also sufficient but not necessary: a work which does nothing at all new will be bad, but a work can do many new things and still leave a bad taste if a homage is handled badly.
It might be helpful to consider traditional Japanese waka poetry as a medium where homage/plagiarism is a much more pervasive and sharper phenomenon: with only 5 lines, a prescribed vocabulary & set of topics, and an extensive corpus & concern for tradition, explicit reuse of lines is necessary, and common. (Indeed, a lot of waka poetry is intelligible and aesthetically appreciable only as allusions, requiring commentary for those of us not reading the originals & steeped in the corpus. Read in isolation, they come off as banal prose observations.) Sharing 1 or 2 lines out of 5 with a previous poem is no flaw, and sharing even 3 may be acceptable. (But at 4, you are in trouble and had better be ready to defend yourself.) What is acceptable? If some important twist or addition or depth can be added by the quotation, or if the remainder of the alluded poem helps explain the current one. The implied gender of the speaker might be reversed, an ‘ending’ provided to an episode, the sentiments flipped into a Buddhist moral, and so on.
Importantly, the new poem relies on the old poem, but the old poem itself gains by the allusion: new possibilities are drawn out of it, or it is given additional resonance by being at the center of a web of associations, so that the thought of it also brings to mind all the others. Good allusions can enrich both works.
EoE and Battle of Okinawa form such a symbiotic pair: Evangelion, like all of Yamato’s progeny, is haunted by WWII and the Japanese Empire (see also Gunbuster’s subtle backgrounding), and the Okinawa allusion places NERV (Japan) being invaded by the UN/SEELE/JSSDF (Americans) using overwhelming force and brutally-efficient methods like flamethrowers and room-by-room grenade clearing and civilian death. Perhaps not a particularly deep set of allusions, but it’s there and serves a purpose, and it also makes watching Battle of Okinawa that much more interesting, as one thinks about how a post-war Japanese film on a defeat is received and transmuted by the youth who grew up in the aftermath and knew only the Japanese economic miracle.
EoE and KlK form less of a pair, and is more commensal. The KlK ending follows EoE beats closely, but with its own textile twist and less of a sekaikei emphasis; knowing how much of Studio Trigger comes from Gainax and the similarity to EoE is… interesting, I suppose? There’s arguably a bit of interesting ‘story archaeology’ in that KlK embraces the original alien panspermia/colonization trope which NGE started with but rapidly pushed into the deep background, but that’s deep into the out-of-universe weeds.
The Flip-Flappers relationship is just parasitic. Flip-Flappers adds nothing to a rewatch of EoE, and indeed, to the extent that someone watches Flip-Flappers and then watches EoE, they suffer from the “Shakespeare is so cliche” problem—EoE will be that much more bland and ordinary. Flip-Flappers, by copying EoE, makes EoE worse off. It is a parasite, worse, one that’s playing a negative-sum game: it poorly benefits itself, at a larger cost to the host.
-
Kemono Friends was a minor phenomenon in 2017: it came out of nowhere from nobody, an ‘adaptation’ of a defunct game, and garnered a fanbase beyond the small children it is ostensibly aimed at, and may be partially responsible for the resurgence of kemonomimi anime. It is much like My Little Pony: Friendship Is Magic in that it is better than it ought to be, due to a staff going above and beyond and surprising even themselves with what they could accomplish on a limited budget—“This anime is the result of giving someone MMD and three dollars”, and much of its strength comes from the simplicity (if you want to be polite; crudeness, if you don’t) of its designs and worldbuilding that hints at much but resolves little, leaving vast empty spaces for the fans to scribble in, knowing that their efforts can match up to the original.
That is not to say that it is actually good, however. The opening theme is catchy (nor is it the only one, how about an Irish rendition of the BGM?), the character designs are remarkably winsome, and the per-episode plots are surprisingly intelligent in depicting the human protagonist as Homo habilis—man the toolmaker and thinker—who lacks wings or claws but can invent new things & ideas to solve problems baffling the animal charaters; but the animation is as laughably bad as claimed (characters routinely just glide over the ground with their walk cycle not making any contact), the worldbuilding is paperthin, and the big finale mostly comes out of nowhere.
The worldbuilding is especially ‘fridge-logic’, as TVTropes puts it: How long do they live? Why are they all female? Do they really just eat machine-made ‘buns’, all of these predator and prey animal friends…? It’s no surprise that one of the more notable fanworks, Abubu’s Arai-san Apartment puts the animal friends in a giant murder-filled SCP/House of Leaves-esque apartment building.
Thus, I’d say that MLP was just better all around. But a more interesting aspect of the comparison is how Kemono Friends is doing since season 1:
It’s dead.. RIP.
A season 2 discarded most of season 1 and the fans loathed it. Followups have been such irrelevancies as a pachinko machine (a traditional way to monetize anything) or another obscure mobile game. No major manga, no movies, no additional seasons, no non-mobile games, no conventions… Sure, there’s still some fanart but that seems to be about it.
What went wrong? Why didn’t KF become a new MLP, Pokemon, or even just a steady-burn subculture like Touhou or Vocaloid?
If Twitter was “a clowncar that fell into a goldmine” then Kadokawa was the clowncar which drove into it and out, complaining all the while about these shiny rocks that keep getting stuck in the tires & who does this guy driving the car think he is anyway?
-
Legend of the Galactic Heroes (Grossly overrated, tedious, and uninsightful.)
LoGH is one of the longest-running and admired anime franchises, still active despite the SF novels starting in 198242ya and the anime adaptation in 1988; this perhaps explains the staggered release of its fully 110 OVA episodes over almost a decade. It is praised for the depths of characterization, intricate strategy, SF worldbuilding, and deep commentary on politics. On paper, LoGH sounds great: a clash of visions as 2 great strategists oppose each other, one the golden child taking over to reform a decayed empire perhaps futilely, and the other the martyred servant of a liberal democracy whose mob leadership repeatedly betrays him out of fear, bringing down on itself the very doom it feared.
Bunk. It is shocking how little happens in 110 deadly-slow episodes, how forgettable the characters are, and how repetitive and shallow the political debates are. I began to fear at the end of the first episode when I realized that the space battle in question consisted of “the outnumbered fleet in the middle will attack the enemy ahead, and then turn around to attack the enemy behind, achieving defeat-in-detail”; it was cribbing from Frederick the Great and his use of interior lines, which is fine—the best mil SF or alt history authors steal shamelessly from history because reality is so much more creative than you are—but it took 2 episodes to achieve something that could’ve been adequately done in half an episode max. (The red flag was when we cut to several officers sitting in a lounge in the middle of a pitched battle sipping wine and calmly noting that it’d been something like 8 hours thus far. In fact, everyone during battles seems awful casual even when their limbs have been blown off and they’re staggering around bleeding out. They must have some wine up there.) The battle animation and explosions are reasonable enough for the era… which is good because you will be seeing a lot of it; several minutes of every episode will just be the same sequence of missiles & explosions in space, copy-pasted again and again to make it ‘epic’. (One really hopes that the sequences are copy-pasted from “the bank” and animators didn’t have to anime thousands of slightly different explosions; given how stretched paper-thin the budget looks overall, I suspect so.) Accelerating to 2× speed helps somewhat. (Saying “it was deliberately an ’80s style” is no excuse; there is plenty of ’80s-style anime which looks great, including lengthy historical dramas like Rose of Versailles.)
The music is passable in a standard ‘90s way. Characters are cardboard, and minimally animated, if perhaps with less reuse than the battles. (Most have one expression; if the writer liked them, they have ’resting cherub face’; if not, ‘resting jerk face’. Main characters will have as many as 2 or 3 expressions.)
The politics are scarcely any better. “An enlightened despot is the best possible ruler!” “But what if he’s followed by a tyrant‽” “Yeah, but what if the people vote in bad presidents‽” “But tyrants!” “But bad votes!” “But tyrants!” (There, I just saved you ~48 hours. Apparently the ‘LoGH’ is short for ‘logorrhea’.) Whole subplots are basically pointless. I still don’t know why so much screentime was devoted to the Earth cult or the ruler of Phezzan, when they could both be deleted with no loss, or why Yang keeps saying “cults never change history” when they do that all the time, and in fact the cult in question does so by killing him (in what is also an extreme Idiot Ball plot twist). I am left wondering if the people praising this as a fantastic space opera have actually read a space opera or mil SF book before, even just something like Honor Harrington.
How about the strategy? I had hopes after the initial episode that we could at least count on retellings of classic battles—why not some battles of Belisarius or Ghengis Khan, but IN SPACE—alas, the author appears to have scrupled at any further plagiarism of history and insisted on originality rather than quality. (All the virtues one detests in an author, and none of the vices one admires.) Thus we are treated to such wonders of strategy of, Lohengrin saying, “there are only 2 routes to the enemy, one through an invincible fortress (the Death Star), and the other through a defenseless neutral planet. What if, instead of attacking the invincible fortress… we invaded the other way?”, and everyone else’s brains promptly exploding at such extraordinary strategic acumen; or the invincible fortress being captured by the stunningly clever ruse of “sending in soldiers dressed in the enemy’s uniform to take it over”. (I’d demand a refund, but, well, you know.)
Worldbuilding? Possibly the worst I have ever seen in mil SF or SF in general. I’m baffled how it’s possible to be this bad at trying to imagine a serious future. There’s no AI of any kind, but that’s a reasonable choice to avoid all the issues that raises, so I’m not complaining about that but much weirder choices. The space warfare is 2D and like Napoleonic era warfare, except when it’s not, and we never get a clear picture of what is possible, or even how the whole space travel is supposed to work: FTL is particularly obscure. How do you have a series about tactics and strategy where the mechanics are so ill-defined? The flaws extend elsewhere. For example, several millennia hence, the Empire is culturally (somehow) a copy of 1700s-era Europe, right down to the wearing of horse-wigs and livery; this has apparently endured for millennia (ie. orders of magnitude longer than the original fashions themselves lasted). This is intended to be literal and no one thinks there’s anything strange about this. OK, somehow a multi-stellar empire of trillions of people manages to stop all social change for thousands of years, despite this contradicting literally all of human history. But it goes on, everything is that bad. What sums it up for me is the third season where our secondary protagonist travels to Earth, countless light-years away, in his starship from the other side of a galactic empire; what does he pull out to earnestly bone up on his Earth history (in what is, incidentally, an exceedingly boring & artless infodump which is the sort of thing which makes ‘infodump’ a dirty word)? Why, what else but a… slide projector, of course! A slide projector. 3,000 years in the future. In an anime made in 199430ya. Look: I was around in 199430ya, and even in 199430ya, we did not see slide projectors that much, and we sure as h—ll did not think slide projectors were going to be a default choice in 3 years, much less 3 millennia. A year after finishing LoGH, it is still coming to mind—a slide projector. How—?
I would like to say something nice about LoGH and come up with something it does well enough to justify its esteem (and 110 episodes) or which at least makes it interesting in some sort of historical or economic way (perhaps as an example of the odd projects that could be funded back in the ’80s–’90s by the OVA business model?), but I can’t. It’s just boring.
-
There’s been plenty of spinoffs—to quote quanticle, “Magia Record is an anime adaptation of a manga adaptation of a JRPG spinoff of… the anime Madoka”—just none have sounded compelling enough for me to bother.↩︎