Join 📚 Josh Beckman's Highlights
A batch of the best highlights from what Josh's read, .
Construal level theory says that we think more concretely, as opposed to abstractly, about things that seem nearer to us in space, time, sociality, chance, and plan. Such concrete thinking can reveal itself, for example, in our using more concrete words as descriptors.
What this work shows is that if you have the right input basic material (data) with the right distribution (here, a heterogeneous one across a bunch of robots), and then you train a high-capacity neural net on it, you get out something greater than the sum of its parts - a model with surprisingly good out-of-distribution generalization as a consequence of some critical reaction that occurs due to your combo of data + architecture + complexity.
Import AI 343: Humanlike AI; LLaMa 2 Protests; The NSA's New AI Center
Jack Clark
More often than not, when we use the language’s primitives, we give up an opportunity to express the domain model of our application. When we use a `String` to represent an `Email`, we haven’t encoded the fact that it needs to satisfy some constraints (ex: have an `@`, or satisfy a regular expression, or have a finite length). When we use `Hash` to represent structured data, we allow the possibility that some keys will be unset, or that some other keys will have values. We can fix all these problems by using classes specific for our use-cases.
Affordance for Errors, Part 2
Guillaume Malette
...catch up on these, and many more highlights
Readwise