Skip to content

Log December 2023


I am writing a bit about each link featured here to provide a bit more value to the reader and also to limit the amount of links I share on this weblog every month.

Mixed topics #

  • Admiral Cloudberg: Extremely detailed analysis of plane crashes and what security measures were implemented after the accident.

  • The Imperfectionist: How to forget what you read: describes "productivity hacks" for being more effective while being less productive - don't stress yourself being productive every single minute of your life and chill out a bit, dude!

  • 1928 in literature: A list of books that become public domain in 2024 - go wild on these! I for one am planning to build an AI that turns all of these into interactive visual novels with the click of a button. Just kidding! (or am I?)

  • Don’t Wake Me Up Too Soon: Very well translated text by the German writer Daniel Kehlman. He writes about the great, unexpected success he had with his books and how to treat your subjects with dignity as an author - especially if those subjects are real people, dead or alive. As well as how fiction is sometimes more believable than reality. Excellent read by a great author - I have read Measuring the world which I sincerely recommend and have Tyll on my reading list.

  • Fast Fashion: Shopping as Entertainment: Blogpost in which the author examines how shopping online is becoming more and more a means to an end. People don't really need the things they are buying in great amounts online - most of them are more or less trash anyway - but they get great pleasure from the act of buying. This is enhanced and gamified by the way online shopping is transformed into an endless casino - like pulling the lever of a one-armed-bandit one is encouraged to buy more and more - each pull of the lever giving a little burst of happiness. Of course that happiness wears off as quickly as it is gained so the next pull of the lever is never far away.

  • The Lion and the Bird: A Tender Illustrated Story About Loneliness, Loyalty, and the Gift of Friendship: I love reading The Marginalian so if I listed every blogpost from them that I like my weblog would be an index of all of their posts. But every once in a while I come across one that I particularly like and this is one of them. The Lion and the bird has everything that makes a great childrens book - it is cute, well illustrated, and talks about the hard topics in life in an understandable and non-cynical way.

  • My Ordinary Life: Improvements Since the 1990s: Similar to The Marginalian gwern consist of mostly great posts - go read it! This one I particularly like since it is a good insulation against the cynicism and nihilism that tends to befall me when I am online too much. gwern makes a giant list of all the things that improved in the last thirty or so years - turns out that is a lot of things!

  • Lock Strategy: A neat guide how to lock your bike securely in dangerous places.

  • The Apollo Syndrome: A great description of a syndrom where teams of highly capable individuals can, collectively, perform badly. In a nutshell a team will perform badly when dealing with prolonged unproductive debates, indecisiveness resulting in omitted tasks, individualistic actions without coordination, and when avoiding confrontation and decision-making.

  • Cloister of Saint-Trophîme by Edouard Baldus: Heavily edited photograph from 1851. An interesting example of how manipulated photos have been around for centuries and not just a creation of the age of Photoshop or image generators.

  • There Once Was an Empire: Blogpost on how a generation of writers perceived the collapse of Austria-Hungary. It seemed to be an eternal empire that would last forever and then the unimaginable happened and it just ended suddenly.

  • Mickey, Disney, and the Public Domain: Mickey or rather the Steamboat Willie version of Mickey is now public domain. This post explores what this means and what you can do now with Mickey

  • Can We Imagine a World Without Work?: Great review of a couple of books that think about if it would be possible to have a world without work. The way I understood it is that there is always a portion of necessary work that we will most likely not be able to automate but that still needs to be done. So there needs to be some system that forces some people to do this work. Capitalism is one of them - Socialism also has punitive systems that force this work upon some people. Also talks about reproductive work - work that does not produce a physical good, but is needed to enable productive work - think of stuff like cooking, cleaning, caring, and so on. Even though this kind of work is not only important but essential for any human society to function it is not paid well and often forced upon women. The text changed my perspective on some of the topics discussed.

  • Sphere Eats the Soul: The author writes about the mindbending hellscape architecture that humans have built in Las Vegas. Just if you thought that place couldn't get any more messed up they went and found a way. Entertaining read on the grotesqueness of modern architecture and just "smartphone bad" in general.

  • The forecasting fallacy: Economy forecasting is consistenly wrong. These handsomely paid economic experts just don't have any idea what will happen in the future - neither with recessions, interest rates, or anything else really. You are better off randomly assuming something than believing in their predictions.

  • Intellectual Collectives Through Use of the Remembrance Agent: Pretty cool idea from the 90s: Put all your text - emails, notes, papers, and so on - into a big database and use that to augment your memory. Seems more feasible than ever with locally run LLMs. I would love to take that idea and build something with a local database and LLM, on my list of things to try out this year for sure.

Almost everything written by Dan Luu is much better than what the average blogger writes and I spent a significant portion of my time recently spent in transit or waiting for transport reading or rereading a lot of his posts. I could probably like dozens here, but that would be kind of moot. I suggest you check out his blog and pick what topics interest you. Check out some of my favourite articles by danluu, more or less software related but also interesting if you are not working in that field:

Software topics #

  • Unfortunately, Kelly Rowland couldn't have used the =HYPERLINK() function to message Nelly: I love this kind of nerd content - taking something obscure way too serious. A great 90s music videos that featured Kelly messaging Nelly with Excel - which would have been technically possible, only Excel was not part of the Nokia 9210 Communicator - which is a great looking product by the way - it was some Nokia software called Sheet 😕
  • The Largest Money-Printing UI Element Ever Made: Blogpost arguing the the browser search bar is the UI element that made the most money ever - because Google will pay you a lot of money to be the default search engine there, even if your browser is relatively obscure and has only fractions of % of marketshare. Write a new browser and get rich wooohooo 🤑!
  • API Tokens: A Tedious Survey: ptacek examines the different strategies to implement API tokens or way to authenticate at an API that he evaluated while building such a system for his startup. Great read for those who want to learn about such strategies or implement ways to authenticate at APIs - to clarify this talks about how to authenticate users as in humans at an API not "inter-API authentication".
  • I am the fold: A cool experiment that shows how dangerous it is to design for "the fold" - the fold is where the browser window ends for a user, so "above the fold" is what you see in your viewport and "below the fold" is where you have to scroll to. Turns out there are many different screen sizes out there and there is no way to pick a correct size.
  • Docs deserve more respect: Docs are important and are often last on the list of priorities. This post gives some good advice how to write good docs. Hire me to write your docs!
  • Donkey Kong: A Record of Struggle: Post on Donkey Kong and the challenges encountered during development. Interesting if you are into oldschool games and the genius workarounds that people came up with to make complicated things work with simple methods back then.
  • Ambient Co-presence: This post talks about a cool idea how to make visitors on a website visible. Cursors are nice, but this sounds less chaotic and better designed.

Software #

  • contrib.rocks: Generate an image of contributors on GitHub for your README.
  • gokrazy: deploy your Go programs as appliances to a Raspberry Pi.
  • google/wuffs: library for parsing, decoding and encoding images, audio, video, fonts and compressed archives.
  • standard-webhooks: Open source tools and guidelines for sending webhooks easily, securely, and reliably
  • tinacms/tinasaurus: Docusaurus + TinaCMS, meaning a relatively simple static site generator using markdown with an easy way to edit for non-technical users.
  • VikParuchuri/marker: converts PDF, EPUB, and MOBI to markdown. Great for turning those huge and pesky formats into a much smaller file format and making it able to template as HTML.
  • xetdata/xetcache: persistent caching of long running functions or jupyter notebook cells. For when I ever have to build more complicated jupyter notebooks again.
  • SOGo | Open Source Groupware: tools for calendar, addresses, and email syncing. I understood it as an OSS replacement for something like Google Workspace or Teams. Not sure how well it works in practice, could not be bothered to test it. There is a demo though (user sogo1, pw sogo)

Imagen topics #

  • Visual Anagrams: Generating Multi-View Optical Illusions with Diffusion Models - pretty impressive technique and many, many different versions of visual anagrams. Some of them are really good, some mid. You can also use this to create jigsaws that can be assembled in two different ways! It does require fiddling to your run yourself - I haven't tested this yet - and also a specific way of prompting. My guess is this will be refined by someone and released as a commercial product - I haven't seen a visual anagram jigsaw yet, so there is some novel product hidden here.
  • cloneofsimo/lora: Loras are the most powerful recent addition to the imagen toolkit - you can finetune a model much faster than with Dreambooth and the result is just a few mb big. Now its feasible to have hundreds of Loras on your local HD and get just as good results as with huge checkpoints that are several gb and take hours to train and finetune.

LLM topics #

  • What I Learned Using Private LLMs to Write an Undergraduate History Essay: Great blogpost in which the author rewrites his old undergraduate history essay using a LLM. I won't spoil it completely but he is able to write the essay faster by a large margin while the overall quality of it is reduced. One thing I found particularly interesting - and it confirms my bias that self-hosted LLMs are the future - is that Ollama perfomed very well. The author presumes that when actually reading the texts that the LLMs cites the result would have been much better - and also less prone to hallucinations of which there were a few - but then again this would have taken much longer as well.
  • List of AI papers
  • StreamDiffusion: An interesting experiment in generating real time video - not sure how usable it is for something serious though - my hardware is a bit too slow.
  • EGjoni/DRUGS: I don't even know what this really does, it just sounded cool. The way I understood it is that we don't change the seed to get different output - so we don't change the noise that is put into the inference step, but somehow change the noise a bit later in the process. The README is unfortunately written by an LLM it seems to kinda hard to make sense of...
  • openchat_3.5: Apparently one of the better chatgpt like models that you can run on your laptop! Try it out with ollama, neato!
  • Mistral "Mixtral" 8x7B 32k model: Another powerful model you can run yourself - you need at least 48GB of RAM though - I have been using it a lot and it compares well to GPT3.5 in my experience.