General Information

Claim Synthesis

Quotes Bin

Double entendre which frames the study of AI safety as “intellectual” (in contrast to War)

“And then there are these additional intellectual questions. Can AI, without robust common sense, be truly safe for humanity?”

double intentre: “intellectual” as in interesting but also “intellectual” as in worth asking

Language of Extremes: AI safety is a “bruce force” problem which requires “extreme scale”

And is brute-force scale really the only way and even the correct way to teach AI? So I’m often asked these days whether it’s even feasible to do any meaningful research without extreme-scale compute.

Quotes Bin

polarization distorts beliefs about others

“Recent years have seen a sharp increase in political polarization in the United States (1–7), leading to deadlock in Congress (8), distorted beliefs about fellow Americans (9, 10), and distrust, hostility, and even violence toward outgroup members (11–13)” [Novoa et al., 2023, p. 1]

generics about particular group is a way that polarized languages manifest

“Specifically, we focus on expressions that make claims about a category as a whole (e.g., “Democrats want to defund the police” makes a claim about the category of “Democrats”), also known as generics (34–38).” [Novoa et al., 2023, p. 2]

One alpha vector per action:

\begin{equation} \alpha^{(k+1)}_{a}(s) = R(s,a) + \gamma \sum_{s’}^{}T(s’|s,a) \max_{a’} \alpha^{(k)}_{a’} (s’) \end{equation}

This is going to give you a set of alpha vectors, one corresponding to each action.

time complexity: \(O(|S|^{2}|A|^{2})\)

you will note we don’t ever actually use anything partially-observable in this. Once we get the alpha vector, we need to use one-step lookahead in POMDP (which does use transitions) to actually turn this alpha vector into a policy, which then does create you

system does not work as well for one type/group of people compared to another

training data really does matter: it may make generalized predictions based on a majority/minority class.

Because IID characteristic of input data, the majority will be over represented

Here’s a PSPACE-COMPLETE language:

\begin{equation} \forall x_1, \exists x_2, \forall x_3 \dots \dots \phi\qty(x_1, x_2, \dots, x_{n}) = 1 \end{equation}