Misuse of Endnotes
I hate when authors uses endnotes for citations and for commentary. Footnotes should be for commentary, endnotes for citations. The problem with mixing the two is then I don't know whether it is worth flipping to the endnotes to see if this footnote says something interesting or if it's just "ibid, p. 64". I'm also in favor of the [Name, Date] inline citation format because if you see the same name come up multiple times you're more likely to go hunt down that reference, but I realize this format can be unwieldy in writing that cites a lot of sources.
Some Cold Water on AI Progress
I did a podcast that just went live, do not read the transcript as I think the AI struggled with my accent.
MBS's Fascination with Alexander the Great
From, Blood and Oil
Mohammed also had an early fascination with Alexander the Great, consuming history books about him and relishing his bold empire building. Some of his closest friends from that time would later refer to Mohammed as their “Iskander,” Arabic for Alexander.
Gustav le Bon on Genius
At the bidding of a Peter the hermit millions of men hurled themselves against the East: the words of an hallucinated enthusiast such as Mahomet created a force capable of triumphing over the Graeco-Roman world; an obscure monk like Luther bathed Europe in blood. The voice of Galileo or a Newton will never have the least echo among the masses. The inventions of genius hasten the march of civilization. The fanatics and the hallucinated create history.
Learning From When Prophecy Fails
When Prophecy Fails is a short book describing a phenomenon that sometimes happens when millenarian predictions fail to take place: believers double down on their beliefs. The book claims there are five conditions that must be present for such increased fervor to occur.
- A belief must be held with deep conviction and it must have some relevance to action, that is, to what the believer does or how he behaves.
- The person holding the belief must have committed himself to it; that is, for the sake of his belief, he must have taken some important action that is difficult to undo. In general, the more important such actions are, and the more difficult they are to undo, the greater is the individual’s commitment to the belief.
- The belief must be sufficiently specific and sufficiently concerned with the real world so that events may unequivocally refute the belief.
- Such undeniable disconfirmatory evidence must occur and must be recognized by the individual holding the belief. [...]
- The individual believer must have social support. It is unlikely that one isolated believer could withstand the kind of disconfirming evidence we have specified. If, however, the believer is a member of a group of convinced persons who can support one another, we would expect the belief to be maintained and the believers to attempt to proselyte or to persuade nonmembers that the belief is correct.
I'm sure most readers can find multiple contemporary examples fitting this pattern.
- Why are modern scientists so dull?...
- ... and are all the non-dull scientists quacks now?
- Methamphetamine production responds to supply and demand.
- Could non-gradient based ML optimization methods find better solutions?
- Quantitative evidence for tipping points and the dictatorship of the small minority.
- "Part of the difficulty of writing scalable and correct multithreaded code is the scarcity of written wisdom from experienced practitioners: oral tradition in lieu of formal writing has left the domain shrouded in mystery. So in the spirit of making this domain less mysterious for our fellow practitioners... we present our collective bag of tricks for writing multithreaded code."
- Neat knob with haptic feedback. I always feel like people are underselling themselves when they call something like this a "hobbyist project". If this were in a real product reviewers would be talking about how cool and innovative it is, the only thing that makes it hobbyist grade is that no one paid him to do it.
- Nice overview of how to design a high performance (~3m msgs/s, 10000 symbol) exchange.
- This video answered a question I have wondered about, does using ReLU as an activation limit what functions a neural network can learn? (The answer is no). Also has an extremely clear description of tensors.
- Some nice looking electronic sculptures.
- One of the more interesting concepts for a modern instrument. Genuinely creative.
- Claims about religion from Peter Turchin.
- Tropical geometry.
- In a previous post I suggested that neural nets could be used for weather forecasting. It looks like I wasn't the only person with this idea.
- Neural nets that are capable of memorizing their inputs still learn to generalize. Why?
- A claim that aging is solved in-vitro.
- Advancing mathematics by guiding human intuition with AI.
- A technical history of AlphaZero.
- A personal history of robotics arms and their programming languages.
- Felt-based rendering. I like these sort of rendering techniques. It reminds me a bit of painterly rendering.
- Nick Patterson on the statistical methods used in the ancient genetics results.
- Some neat cross-sectional photography of electronic components.
- How the empire could have won. I've often thought the Straussian interpretation of Star Wars is that the empire are the good guys.
- A prediction that human genome-editing will be commonly available in 10-20 years, and some speculative discussion of the consequences of that. This timeline seems wildly optimistic to me. I am willing to bet a substantial amount against this occurring at the speed claimed.
- LASSO can be implemented as a simple neural network.
- Solving PDE's without FEM using Monte Carlo Geometry Processing.
- Excellent visual overview of different types of splines.
- LR and recursive descent are the only two parsing techniques most people need to know.
- Some speculation about extraterrestrial life.