--- title: October 2021 News description: October 2021 Gwern.net newsletter with links on TODO created: 2020-01-02 status: in progress previous: /newsletter/2021/09 next: /newsletter/2021/11 confidence: log cssExtension: dropcaps-de-zs backlink: False ... October 2021's [Gwern.net](https://gwern.net/newsletter/2021/10) [newsletter](https://gwern.substack.com/ "'Gwern.net newsletter (Substack subscription page)', Branwen 2013") is now out; previous, [September 2021](https://gwern.net/newsletter/2021/09) ([archives](https://gwern.net/doc/newsletter/index)). This is a collation of links and summary of major changes, overlapping with my [Changelog](https://gwern.net/changelog); brought to you by my donors on [Patreon](https://www.patreon.com/gwern). # Writings **Gwern.net**, interesting new tags:
- [AlphaFold](https://gwern.net/doc/ai/nn/transformer/alphafold/index) - [OA CLIP](https://gwern.net/doc/ai/nn/transformer/clip/index) - [DALL·E](https://gwern.net/doc/ai/nn/transformer/gpt/dall-e/index) - [Jukebox](https://gwern.net/doc/ai/nn/transformer/gpt/jukebox/index) - [model-based DRL](https://gwern.net/doc/reinforcement-learning/model/index): - [A0](https://gwern.net/doc/reinforcement-learning/model/alphago/index)/[MuZero](https://gwern.net/doc/reinforcement-learning/model/muzero/index) - [chess AI](https://gwern.net/doc/reinforcement-learning/chess/index), [psychology](https://gwern.net/doc/psychology/chess/index) - [Decision Transformer](https://gwern.net/doc/reinforcement-learning/model/decision-transformer/index) - [MARL](https://gwern.net/doc/reinforcement-learning/multi-agent/index)/[imperfect-information](https://gwern.net/doc/reinforcement-learning/imperfect-information/index): - [poker](https://gwern.net/doc/reinforcement-learning/imperfect-information/poker/index), [_Diplomacy_](https://gwern.net/doc/reinforcement-learning/imperfect-information/diplomacy/index), [_Hanabi_](https://gwern.net/doc/reinforcement-learning/imperfect-information/hanabi/index) - [model-free DRL](https://gwern.net/doc/reinforcement-learning/model/decision-transformer/index) - [OA5](https://gwern.net/doc/reinforcement-learning/model-free/oa5/index)/[AlphaStar](https://gwern.net/doc/reinforcement-learning/model-free/alphastar/index) - [exploration in DRL](https://gwern.net/doc/reinforcement-learning/exploration/index) - [robotics](https://gwern.net/doc/reinforcement-learning/robot/index) - [adversarial examples](https://gwern.net/doc/ai/nn/adversarial/index) - [mixture-of-experts](https://gwern.net/doc/ai/scaling/mixture-of-experts/index) - [emergence](https://gwern.net/doc/ai/scaling/emergence/index) - [diffusion generative models](https://gwern.net/doc/ai/nn/diffusion/index) - [video generation](https://gwern.net/doc/ai/video/generation/index)/[analysis](https://gwern.net/doc/ai/video/analysis/index) - [RNN](https://gwern.net/doc/ai/nn/rnn/index)/[Transformer](https://gwern.net/doc/ai/nn/transformer/index)/[MLP](https://gwern.net/doc/ai/nn/fully-connected/index) - [BigGAN](https://gwern.net/doc/ai/nn/gan/biggan/index) - [Codex/Copilot](https://gwern.net/doc/ai/nn/transformer/gpt/codex/index) - [text model 'inner monologue'](https://gwern.net/doc/ai/nn/transformer/gpt/inner-monologue/index) - [text style transfer](https://gwern.net/doc/ai/text-style-transfer/index) - [tabular ML](https://gwern.net/doc/ai/tabular/index) - ['brain imitation learning'](https://gwern.net/doc/reinforcement-learning/imitation-learning/brain-imitation-learning/index) - [meta reinforcement learning](https://gwern.net/doc/reinforcement-learning/meta-learning/index) - [preference learning](https://gwern.net/doc/reinforcement-learning/preference-learning/index) - [AI safety](https://gwern.net/doc/reinforcement-learning/safe/index) - [experience curves](https://gwern.net/doc/economics/experience-curve/index) - [text games](https://gwern.net/doc/fiction/text-game/index) - [music](https://gwern.net/doc/music/index) - [cellular automata](https://gwern.net/doc/cs/cellular-automaton/index) - [cloning](https://gwern.net/doc/genetics/cloning/index) - [gametogenesis](https://gwern.net/doc/genetics/gametogenesis/index) - [genome synthesis](https://gwern.net/doc/genetics/genome-synthesis/index) - [adoption studies](https://gwern.net/doc/genetics/heritable/adoption/index) - [rare variant genetics](https://gwern.net/doc/genetics/heritable/rare/index) - [genome sequencing](https://gwern.net/doc/genetics/sequencing/index) - [the microbiome](https://gwern.net/doc/genetics/microbiome/index) - [index selection](https://gwern.net/doc/genetics/selection/artificial/index-selection/index) - [intermittent fasting/CR](https://gwern.net/doc/longevity/fasting/index) - [semaglutide](https://gwern.net/doc/longevity/glp/semaglutide/index)/[tirzepatide](https://gwern.net/doc/longevity/glp/tirzepatide/index)/ - [epigenetic reprogramming/clocks](https://gwern.net/doc/longevity/epigenetics/index) - [senolytics](https://gwern.net/doc/longevity/senolytic/index) - [marijuana](https://gwern.net/doc/marijuana/index) - [psychedelics](https://gwern.net/doc/psychedelic/index) - [meditation](https://gwern.net/doc/psychiatry/meditation/index) - [psychological depth illusion](https://gwern.net/doc/psychology/cognitive-bias/illusion-of-depth/index) - [inner monologue](https://gwern.net/doc/psychology/inner-voice/index) - [inverted U-curve of novelty](https://gwern.net/doc/psychology/novelty/index) - [human olfaction](https://gwern.net/doc/psychology/smell/index) - [willpower](https://gwern.net/doc/psychology/willpower/index)/[energy](https://gwern.net/doc/psychology/energy/index) - [short sleepers](https://gwern.net/doc/zeo/short-sleeper/index) - [Dogs](https://gwern.net/doc/dog/index): [genetics](https://gwern.net/doc/genetics/heritable/dog/index), [cloning](https://gwern.net/doc/genetics/cloning/dog/index) - [preference falsification](https://gwern.net/doc/sociology/preference-falsification/index) - [survival analysis](https://gwern.net/doc/statistics/survival-analysis/index) - [cryonics](https://gwern.net/doc/cryonics/index) - [carbon capture tech](https://gwern.net/doc/technology/carbon-capture/index) - [nuclear subterrenes/self-sinking waste](https://gwern.net/doc/technology/self-sinking/index) - [The Digital Antiquarian](https://gwern.net/doc/technology/digital-antiquarian/index) - [The Public Domain Review](https://gwern.net/doc/history/public-domain-review/index) - [data visualization](https://gwern.net/doc/design/visualization/index) - [typography](https://gwern.net/doc/design/typography/index) - [crime](https://gwern.net/doc/crime/index)
# Links ## AI - ["Magnetic control of tokamak plasmas through deep reinforcement learning"](https://www.nature.com/articles/s41586-021-04301-9#deepmind), Degrave et al 2022 [Matters Of](https://www.reddit.com/r/mlscaling/ "'ML Scaling subreddit', Branwen 2020") [Scale](https://gwern.net/note/scaling): - ["Program Synthesis with Large Language Models"](https://arxiv.org/abs/2108.07732#google), Austin et al 2021 (smooth scaling like [Codex](https://arxiv.org/abs/2107.03374#openai "'Evaluating Large Language Models Trained on Code', Chen et al 2021") continues to 137b-parameter [LaMDA](https://blog.google/technology/ai/lamda/); enables dialogue, solving math problems); ["Show Your Work: Scratchpads for Intermediate Computation with Language Models"](https://arxiv.org/abs/2112.00114#google), Anonymous et al 2021 ([AQUA-RAT](https://arxiv.org/abs/1705.04146#deepmind "'Program Induction by Rationale Generation : Learning to Solve and Explain Algebraic Word Problems', Ling et al 2017") might be worth revisiting) - ["The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Models"](https://arxiv.org/abs/2201.03544), Pan et al 2022 ("phase transitions: capability thresholds at which the agent's behavior qualitatively shifts") - ["CoCa: Contrastive Captioners are Image-Text Foundation Models"](https://arxiv.org/abs/2205.01917#google "‘CoCa: Contrastive Captioners are Image-Text Foundation Models’, Yu et al 2022"), Yu et al 2022 (86% ImageNet zero-shot) - [CLIPIT PixelDraw](https://twitter.com/dribnet/status/1427613617973653505 "Here's a link to my colab if you'd like to give it a go yourself. This codebase builds off of previous work from many people including @advadnoun @RiversHaveWings @NerdyRodent as well as ClipDraw from @kvfrans @crosslabstokyo @err_more and @okw") ([Colab](https://colab.research.google.com/github/dribnet/clipit/blob/master/demos/PixelDrawer.ipynb "CLIPIT PixelDraw demo"); high-quality pixel art via CLIP which Just Works™, using the [CLIPDraw](https://arxiv.org/abs/2106.14843 "'CLIPDraw: Exploring Text-to-Drawing Synthesis through Language-Image Encoders', Frans et al 2021") approach; this is particularly striking given the failure of pixel art generative models to date, like the endless Pokemon sprite attempts---apparently it really did need _n_ = 400m before pixel art levels of visual abstraction could work... See also ["Projected GANs Converge Faster", Sauer et al 2021](https://arxiv.org/abs/2111.01007 "'Projected GANs Converge Faster', Sauer et al 2021") which gets its good Pokemon via ImageNet/CLIP) - ["Make Every feature Binary (MEB): A 135b-parameter sparse neural network for massively improved search relevance"](https://www.microsoft.com/en-us/research/blog/make-every-feature-binary-a-135b-parameter-sparse-neural-network-for-massively-improved-search-relevance/) (an extremely large embedding for text ranking, now powering 100% of Microsoft Bing search queries) - ["Partial success in closing the gap between human and machine vision"](https://arxiv.org/abs/2106.07411), Geirhos et al 2021 ("We have to admit that we view our results concerning the benefits of increasing dataset size by 1--3 orders of magnitude with mixed feelings. On the one hand, 'simply' training standard models on (a lot) more data certainly has an intellectually disappointing element—particularly given many rich ideas in the cognitive science and neuroscience literature..."); ["Evaluating Machine Accuracy on ImageNet"](https://openreview.net/forum?id=Q3R088EFtng#google), Shankaar et al 2021 ("the latest models from 2020 are on par with our best human labeler"); ["When does dough become a bagel? Analyzing the remaining mistakes on ImageNet"](https://arxiv.org/abs/2205.04596#google), Vasudevan et al 2022 ("[CoCa-FT](https://arxiv.org/abs/2205.01917#google "‘CoCa: Contrastive Captioners are Image-Text Foundation Models’, Yu et al 2022") gets 42 of the 68 [remaining hard errors] correct") - ["What Changes Can Large-scale Language Models Bring? Intensive Study on HyperCLOVA: Billions-scale Korean Generative Pretrained Transformers"](https://arxiv.org/abs/2109.04650#naver), Kim et al 2021 ([Naver](https://en.wikipedia.org/wiki/Naver)) - [DeepMind 2020 budget](https://gwern.net/doc/reinforcement-learning/deepmind/2020-deepmind-fullaccounts.pdf#page=3) reaches \$1b (\$1.06b, +\$0.085b) - ["M6-10T: A Sharing-Delinking Paradigm for Efficient Multi-Trillion Parameter Pretraining"](https://arxiv.org/abs/2110.03888#alibaba), Anonymous 2021 (easy GPT-10t training via weight-tying: pretrain a single module/layer that fits in nodes, then 'unroll' it to the fullsize model for finetuning) - ["Persia: A Hybrid System Scaling Deep Learning Based Recommenders up to 100 Trillion Parameters"](https://arxiv.org/abs/2111.05897 "‘Persia: An Open, Hybrid System Scaling Deep Learning-based Recommenders up to 100 Trillion Parameters’, Lian et al 2021"), Lian et al 2021 ([Kuaisho](https://en.wikipedia.org/wiki/Kuaishou)) ## Genetics Everything Is Heritable: Recent Evolution: Engineering: ## Statistics/Meta-Science ## Politics/Religion ## Psychology/Biology - ["Adolescent cannabis use and adult psychoticism: A longitudinal co-twin control analysis using data from two cohorts"](https://gwern.net/doc/marijuana/2021-schaefer.pdf), Schaefer et al 2021 ("no") - [Clonally transmissible cancer](https://en.wikipedia.org/wiki/Clonally_transmissible_cancer) - ["How Long Does It Take Ordinary People To 'Get Good' At Chess?"](https://github.com/jcw024/lichess_database_ETL/blob/main/README.md), Joseph Wong - ["Did we find a copycat? 'Do as I Do' in a domestic cat (_Felis catus_)"](https://link.springer.com/article/10.1007/s10071-020-01428-6 "'Did we find a copycat? ‘Do as I Do’ in a domestic cat (Felis catus)', Fugazza et al 2020"), Fugazza et al 2020 - ["Speed-accuracy trade-off in plants"](https://link.springer.com/article/10.3758/s13423-020-01753-4), Ceccarini et al 2020 - ["Wet mammals shake at tuned frequencies to dry"](https://royalsocietypublishing.org/doi/10.1098/rsif.2012.0429), Dickerson et al 2021 ([Sandberg thread](https://twitter.com/anderssandberg/status/1449151410575130627 "#FridayPhysicsFun — I got a question from the ever awesome @Gwern: shaking off water from spoons and dishes, does it actually dry them faster than just letting it drip in constant gravity? What about dropping them?")) - [Human echolocation](https://en.wikipedia.org/wiki/Human_echolocation) - [Impossible colors](https://en.wikipedia.org/wiki/Impossible_color) ("Hello, yes, I would like a [blacker black](https://en.wikipedia.org/wiki/Rich_black), also, a ['self-luminous red'](https://en.wikipedia.org/wiki/Impossible_color#Chimerical_colors) which is simultaneously red & brighter than white"---[real colors]{.smallcaps}, from _the demented ravings of **madmen**_; [THEY HAVE PLAYED US FOR FOOLS!]{.smallcaps}) - ["Direct voluntary control of pupil constriction and dilation: Exploratory evidence from pupillometry, optometry, skin conductance, perception, and functional MRI"](https://www.sciencedirect.com/science/article/pii/S0167876021008448), Eberhardt et al 2021 ## Technology - [History of Air Guns](https://en.wikipedia.org/wiki/Air_gun#History) - ["AIR-FI: Generating Covert Wi-Fi Signals from Air-Gapped Computers"](https://arxiv.org/abs/2012.06884), Guri 2020 (some fun earlier side-channels: [Guri et al 2015a](https://www.usenix.org/system/files/sec15-paper-guri-update_v2.pdf#page=2 "GSMem: Data Exfiltration from Air-Gapped Computers over GSM Frequencies"), [Guri et al 2015b](https://ieeexplore.ieee.org/document/7243739/ "BitWhisper: Covert Signaling Channel between Air-Gapped Computers Using Thermal Manipulations")) - ["Syntax highlighting is a waste of an information channel"](https://buttondown.email/hillelwayne/archive/syntax-highlighting-is-a-waste-of-an-information/) - ["Ditherpunk---The article I wish I had about monochrome image dithering"](https://surma.dev/things/ditherpunk/) - **Block writing**: - [Ed Interlock](https://fontsinuse.com/typefaces/40498/ed-interlock) - Anita Jürgeleit's 2016 ["Hangulatin"](https://www.myfonts.com/collections/hangulatin-en-font-urw): a [Hangul](https://en.wikipedia.org/wiki/Hangul)-inspired font for English/German word blocks ; Sebastian Moser's ["Sori"](http://luc.devroye.org/fonts-78250.html) - [Nihon Typeface](https://www.designmadeingermany.de/2014/61882/) - [Xu Bing's](https://en.wikipedia.org/wiki/Xu_Bing) ["Square Word Calligraphy"](https://twitter.com/incunabula/status/1573222769978114048) ([Omniglot](https://www.omniglot.com/conscripts/swc.htm)) - [Square Kufic](https://en.wikipedia.org/wiki/Kufic#Square_Kufic) (eg. [ISIS](https://commons.wikimedia.org/wiki/File:ISIL_Caliphate_Seal.svg)) - [Calcula](https://www.typotheque.com/articles/calcula) ([review](https://typographica.org/typeface-reviews/calcula/)) - [Cyrillic _Vyaz_](https://en.wikipedia.org/wiki/Vyaz_(Cyrillic_calligraphy)) ## Economics - ["What is the Commons Worth? Estimating the Value of Wikimedia Imagery by Observing Downstream Use"](https://gwern.net/doc/economics/copyright/2018-erickson.pdf), Erickson et al 2018; ["Digitization and the Demand for Physical Works: Evidence from the Google Books Project"](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3339524), Nagaraj & Reimers 2019 ([copyright deadweight losses](https://gwern.net/doc/economics/copyright/index)) - ["Potterian Economics"](https://gwern.net/doc/economics/2017-levy.pdf), Levy & Snir 2017 - ["The Theory of Interstellar Trade"](https://gwern.net/doc/economics/2010-krugman.pdf), Krugman 1978/2010 - ["Should subscription-based content creators display their earnings on crowdfunding platforms? Evidence from Patreon"](https://gwern.net/doc/economics/2021-crosby.pdf), Crosby & McKenzie 2021 (I'm giving disabling my public display a try... Doesn't seem to be working thus far.) ## Philosophy ## Fiction ## Miscellaneous # Books **Nonfiction**: **Fiction**: # Film/TV **Live-action**: **Animated**: # Music [**Touhou**](https://www.reddit.com/r/TOUHOUMUSIC/search/?q=author%3Agwern&sort=new&restrict_sr=on&t=all): [**Vocaloid**](https://www.reddit.com/r/Vocaloid/search/?q=author%3Agwern&sort=new&restrict_sr=on&t=all): **MLP**: **Doujin**: **Misc**: