CS Lewis — How will the bomb find you?

In one way we think a great deal too much of the atomic bomb. ‘How are we to live in an atomic age?’ I am tempted to reply: ‘Why, as you would have lived in the sixteenth century when the plague visited London almost every year […]; or indeed, as you are already living in an age of cancer, an age of syphilis, an age of paralysis, an age of air raids, an age of railway accidents, an age of motor accidents. In other words, do not let us begin by exaggerating the novelty of our situation. Believe me, dear sir or madam, you and all whom you love were already sentenced to death before the atomic bomb was invented… It is perfectly ridiculous to go about whimpering and drawing long faces because the scientists have added one more chance of painful and premature death to a world which already bristled with such chances and in which death itself was not a chance at all, but a certainty. […] If we are all going to be destroyed by an atomic bomb, let that bomb when it comes find us doing sensible and human things—praying, working, teaching, reading, listening to music, bathing the children, playing tennis, chatting to our friends over a pint and a game of darts—not huddled together like frightened sheep and thinking about bombs.

The Weather Year Round Anywhere on Earth - Weather Spark

Relational E-Matching

We present a new approach to e-matching based on relational join; in particular, we apply recent database query execution techniques to guarantee worst-case optimal run time. Compared to the conventional backtracking approach that always searches the e-graph “top down”, our new relational e-matching approach can better exploit pattern structure by searching the e-graph according to an optimized query plan. We also establish the first data complexity result for e-matching, bounding run time as a function of the e-graph size and output size. We prototyped and evaluated our technique in the state-of-the-art egg e-graph framework. Compared to a conventional baseline, relational e-matching is simpler to implement and orders of magnitude faster in practice.

Intransitive dice - Wikipedia

  • Die /A/ has sides 2, 2, 4, 4, 9, 9.
  • Die /B/ has sides 1, 1, 6, 6, 8, 8.
  • Die /C/ has sides 3, 3, 5, 5, 7, 7.

    The probability that /A/ rolls a higher number than /B/, the probability that /B/ rolls higher than /C/, and the probability that /C/ rolls higher than /A/ are all 5/9

Corporate Memphis - Wikipedia

Visual Programming Codex: Some random App Store illustration

Improving the New York Times’ line wrap balancer

What every software engineer should know about search

PlayStation 3 Architecture | A Practical Analysis

Typed Image-based Programming with Structure Editing

18000 people in a single building. (Saint Petersburg, Russia) : UrbanHell

Best Copycat Levain Cookie Recipe - The Pancake Princess #recipes

Compares nine recipes

Capitol Hill Babysitting Co-op - Wikipedia

The Capitol Hill Babysitting Cooperative (CHBC) is a cooperative located in Washington, D.C. , whose purpose is to fairly distribute the responsibility of babysitting between its members. The co-op is often used as an allegory for a demand -oriented model of an economy. The allegory illustrates several economic concepts, including the paradox of thrift and the importance of the money supply to an economy’s well-being. At first, new members of the co-op felt, on average, that they should save more scrip before they began spending. So they babysat whenever the opportunity arose, but did not spend the scrip they acquired. Since babysitting opportunities only arise when other couples want to go out, there was a shortage of demand for babysitting. This illustrates the phenomenon known as the paradox of thrift . The administration’s initial reaction to the co-op’s recession was to add new rules. But the measures did not resolve the inadequate demand for babysitting. Eventually, the co-op was able to alleviate the issue by giving new members thirty hours’ worth of scrip, but only requiring them to return twenty when they left the co-op. Within a few years a new problem arose. There was too much scrip and a shortage of babysitting. As new members joined, more scrip was added to the system until couples had too much, but new members were not able to spend it because no one else wanted to babysit. In general, the cooperative experienced regular problems because the administration took in more than it spent, and at times the system added too much scrip into the system via the amount issued to new members.

Google API Improvement Proposals

AIPs are design documents that summarize Google’s API design decisions. They also provide a framework and system for others to document their own API design rules and practices.

“Dune” (the movie), annotated - by Max Read

Culture Shock - by Siddhesh - Obvious Bicycle

Impressions of an Indian graduate student new to America

Curves and Surfaces – Bartosz Ciechanowski

An oral history of Bank Python

UAX #31: Unicode Identifier and Pattern Syntax

/This annex describes specifications for recommended defaults for the use of Unicode in the definitions of general-purpose identifiers, immutable identifiers, hashtag identifiers, and in pattern-based syntax. It also supplies guidelines for use of normalization with identifiers./

Median voter theorem - Wikipedia

It states that if voters and policies are distributed along a one-dimensional spectrum , with voters ranking alternatives in order of proximity, then any voting method which satisfies the Condorcet criterion will elect the candidate closest to the median voter. In particular, a majority vote between two options will do so.

Dirac’s belt trick, Topology, and Spin ½ particles

Excellent video. Especially focuses on SO(3) and SU(2).

Padé Approximants

Like Taylor Series but better behaved for functions that don’t go to infinity. Goes to 0 instead. Often stays close to the function for longer.

Steph Curry 105 THREES IN A ROW, 5 minutes straight without missing

Discussion with Eliezer Yudkowsky on AGI interventions

The following is a partially redacted and lightly edited transcript of a chat conversation about AGI between Eliezer Yudkowsky and a set of invitees in early September 2021. By default, all other participants are anonymized as “Anonymous”.

Links from Steve Omohundro:

  • Interval Methods A collection of links to interval methods
  • Scientific Contributions | Steve Omohundro (see “bump trees”) “Steve developed a wide range of algorithms that dramatically speedup neural computations on ordinary serial computers. He introduced numerous data structures based on kd-trees with various models at the leaves to implement a wide range of statistical and learning tasks. He introduced the boxtree, balltree, octboxtree, and bumptree data structures for speeding up geometric algorithms and algorithms based on mixtures of Gaussians.”
  • 1701.06538 Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer “They’re also pretty terrible for learning since most weights don’t need to be updated for most training examples and yet they are. Google and others are using Mixture-of-Experts to avoid some of that cost”
  • 2106.10860 Multiplying Matrices Without Multiplying “Matrix multiply is a pretty inefficient primitive and alternatives are being explored”
  • Probabilistic Circuits: Representations, Inference, Learning and Theory “Here’s a nice 3 hour long tutorial about “probabilistic circuits” which is a representation of probability distributions, learning, Bayesian inference, etc. which has much better properties than most of the standard representations used in statistics, machine learning, neural nets, etc. It looks especially amenable to interpretability, formal specification, and proofs of properties.”

I was halfway through a PhD on software testing and verification before joining Anthropic (opinions my own, etc), and I’m /less/ convinced than Eliezer about theorem-proving for AGI safety.

I want to push back against the idea that ANNs are “vectors of floating points” and therefore it’s impossible to prove things about them. Many algorithms involve continuous variables and we can prove things about them. Support vector machines are also learning algorithms that are “vectors of floating points” and we have a pretty good theory of how they work. In fact, there already is a sizable body of theoretical results about ANNs, even if it still falls significantly short of what we need. The biggest problem is not necessarily in the “floating points”. The problem is that we still don’t have satisfactory models of what an “agent” is and what it means for an agent to be “aligned”. But, we do have some leads. And once we solve this part, there’s no reason of principle why it cannot be combined with some (hitherto unknown) theory of generalization bounds for ANNs.

this post seems to argue (reasonably convincingly, in my view) that the space of possible abstractions (“epistemic representations”) is discrete rather than continuous, such that any representation of reality sufficiently close to “human compressions” would in fact /be/ using those human compressions, rather than an arbitrarily similar set of representations that comes apart in the limit of strong optimization pressure

The Value of Nothing: Capital versus Growth - American Affairs Journal

“Of course, Andreessen was correct to claim that software was eating the world, but he had the causation backwards. Software’s high valuations were not the result of its extraordinary technological promise. Rather, the software sector had become the primary locus of innovation because of its high valuations. Its financial characteristics allowed software to attract growth investment while other sectors no longer could.

Twitter