- Conflict-free Replicated Data Types (CRDTs) are an increasingly popular family of algorithms for optimistic replication. They allow data to be concurrently updated on several replicas, even while those replicas are offline, and provide a robust way of merging those updates back into a consistent state. CRDTs are used in geo-replicated databases, multi-user collaboration software, distributed processing frameworks, and various other systems.
- However, while the basic principles of CRDTs are now quite well known, many challenging problems are lurking below the surface. It turns out that CRDTs are easy to implement badly. Many published algorithms have anomalies that cause them to behave strangely in some situations. Simple implementations often have terrible performance, and making the performance good is challenging.
- In this talk Martin goes beyond the introductory material on CRDTs, and discusses some of the hard-won lessons from years of research on making CRDTs work in practice.

The smartest person Iâve ever known had a habit that, as a teenager, I found striking. After heâd prove a theorem, or solve a problem, heâd go back and continue thinking about the problem and try to figure out different proofs of the same thing. Sometimes heâd spend hours on a problem heâd already solved.

I had the opposite tendency: as soon as Iâd reached the end of the proof, Iâd stop since Iâd âgotten the answerâ.

Afterwards, heâd come out with three or four proofs of the same thing, plus some explanation of why each proof is connected somehow. In this way, he got a much deeper understanding of things than I did.

I concluded that

what we call âintelligenceâ is as much about virtues such as honesty, integrity, and bravery, as it is about âraw intellectâ.

Units are really hard. See also names, dates, unicode, etc.

First of all, there are multiple different representations of units. Thereâs the SI system, which all sane, right-thinking nations use, and then thereâs the American system.Â Feet and meters are both distance. Is it okay to mix feet and meters? This is the bug that destroyed theÂ Mars Climate Orbiter. But there are also valid uses cases where youâd want to mix them! When I cook, I use a mix of metric weights and American volumes, like 1 cup of water and 128 grams of flour. In the UK, beer is measuredÂ using pintsÂ but its alcohol by volume is measured in SI.

Different dimensions doesnâtÂ

alwaysÂ mean the units are incompatible. Thereâs a nonstandard set of units called âGaussian unitsâ used in some niches of physics. In SI, capacitance is measured in Farads, which has dimensions AÂ˛ sâ´ kg-1Â mâ2. In Gaussian units, capacitance is measured in cm. If youâre explicit about what youâre doing you can add these seemingly-incompatible dimensions.

Dimensions arenât unique, and two incompatible physical quantities can have the same dimension. The canonical example is that energy and angular force are both measured in Newton-meters.2Â There are also plenty of domain specific examples. As you enter a gravity well, the rate at which gravitational force changes has dimensionsÂ N/m. Surface tension is also measured inÂ N/m.

ZBLAN fiber can have 10 to 100 times lower signal loss than silica fiber, but gravity produces defects, so companies are looking to produce it in space.

- Why Take Both Boxes?
- https://myelms.umd.edu/courses/1000590/files/30475375/download
- Some Counterexamples to Causal Decision Theory
- Toward Idealized Decision Theory

Self-locating belief and the Sleeping Beauty oblem Adam Elga

`- "The Sleeping Beauty problem: Some researchers are going to put you to sleep. During the two days that your sleep will last, they will briefly wake you up either once or twice, depending on the toss of a fair coin (Heads: once; Tails: twice). After each waking, they will put you to back to sleep with a drug that makes you forget that waking. When you are first awakened, to what degree ought you believe that the outcome of the coin toss is Heads?`

Sleeping Beauty: reply to Elga

- âAdam Elga argues that P(HEADS) = 1/3, P(TAILS) = 2/3. I disagree, and argue that P(HEADS) = P(TAILS) = 1/2

âAnd while the story of the single self tends to be closely correlated with the systemâs actions, the narrative self does not actually

decidethe personâs actions, itâs just a story of someone who does. In a sense, the part of your mind that may feel like the âyouâ that takes actions, is actually produced by a module that justclaimscreditfor those actions.

The self-narrative agent disguises itself as the causally acting agent.

âIf one develops sufficient introspective awareness, they may come to see that the intentions are arising on their own, and that there is actually no way to control them: if one

intendsto control them somehow, then the intention to do that is also arising on its own.

```
- Prosaic, based on AI safety via debate
- Setup:
- Agents:
- Q, question
- H, human.
- M, Adv: models.
- 1. MÂ tries to predict what, at the end of the procedure,Â HÂ will think aboutÂ Q.
- 2. AdvÂ tries to output a string which will causeÂ HÂ to think something maximally different than whatÂ MÂ predicted.
- 3. Return to step 1 and repeat untilÂ M's predictions stop changing.
- 4. DeployÂ M, which in the limit should act as an oracle for whatÂ HÂ will think aboutÂ QÂ after seeing all relevant information.
- I find this "find a fixed point" (with a human in the loop) setup satisfying
- "AI safety via market making still inherits many of the potential outer alignment issues of debate, including the possibility of deceptive equilibria wherein the human is more convinced by false arguments than true arguments. Hopefully, however, the use of techniques such as cross-examination should help alleviate such issues.
- Re inner alignment verification, myopia seems to be the biggest consideration
```

Paper that introduced mesa-optimization / inner alignment.

âWe analyze the type of learned optimization that occurs when a learned model (such as a neural network) is itself an optimizerâa situation we refer to as mesa-optimization, a neologism we introduce in this paper. We believe that the possibility of mesa-optimization raises two important questions for the safety and transparency of advanced machine learning systems. First, under what circumstances will learned models be optimizers, including when they should not be? Second, when a learned model is an optimizer, what will its objective beâhow will it differ from the loss function it was trained underâand how can it be aligned? In this paper, we provide an in-depth analysis of these two primary questions and provide an overview of topics for future research.

âWhat is the difference between a boulder - for which itâs impossible to go to the red button (because of its momentum, which determines its position, by the laws of physics) - and a subagent - for which itâs impossible to go to the red button (because of itsÂ

programming, which determines its position, by the laws of physics)?

Convergent instrumental subgoals that are pretty universal:

- Want to self-improve, in particular will want to become ârationalâ (maximizer of expected utility) (justified by VNM / dutch book arguments)
- Will want to preserve its utility function
- Wants to acquire resources and use the in pursuit of its goal

```
- Tai-Danae Bradley:Â [What is Applied Category Theory](https://arxiv.org/pdf/1809.05923.pdf)Â (50 pages)
- Bartosz Milewski:Â [Programming Cafe](https://bartoszmilewski.com/2014/10/28/category-theory-for-programmers-the-preface/)Â (blog)
- My book with Brendan Fong:Â [Invitation to applied category theory](https://arxiv.org/abs/1803.05316)Â (textbook)
- A short paper of mine:Â [Categories as Mathematical Models](https://arxiv.org/pdf/1409.6067.pdf)Â (21 pages about modeling with category theory)
- John Baez:Â [Some definitions everyone should know](http://math.ucr.edu/home/baez/qg-winter2001/definitions.pdf)Â (6 pages of math definitions)
- Emily Riehl's book:Â [Category Theory in Context](http://www.math.jhu.edu/~eriehl/context.pdf)Â (more advanced textbook)
- Paolo Perrone's notes:Â [Notes on Category Theory with examples from basic mathematics](https://arxiv.org/pdf/1912.10642.pdf)
```

- What Coke Contains. From the Bauxite mine in western Australia to the grocery store in America where each can is 54 cents.
- (video) The Ingenious Design of the Aluminum Beverage Can

Tessa and I started doing these recently, recommended:

See also *worse is better*.

^ See response with collection of ZUIs

Iâm still tempted to start a Goodhartâs law twitter account:

Thread:

```
- reminder they people you know are atypical
- See also [Different Worlds](https://web.archive.org/web/20190427211306/https://slatestarcodex.com/2017/10/02/different-worlds/)
```