I asked for further details on the 10th point and Claude listed a bunch of stuff I’ve absolutely never heard of. I’d say it’s probably related to meditation if I had to guess. Here’s that.
—
Claude Let me break down Time Perception Management into its deeper components, as this is one of the most subtle yet powerful micro-skills...
This skill set is particularly valuable because most people are unconscious of how their time perception affects their decisions and experiences. Mastering it gives you significant advantages in:
Would you like me to elaborate on any particular aspect? Or shall we explore how this connects with some of the other micro-skills we discussed?
alfred-harwood on Abstractions are not NaturalYour reaction seems fair, thanks for your thoughts! Its a good a suggestion to add an epistemic status - I'll be sure to add one next time I write something like this.
andrew-mcknight on Against Almost Every Theory of Impact of InterpretabilityDo you think putting extra effort into learning about existing empirical work while doing conceptual work would be sufficient for good conceptual work or do you think people need to be producing empirical work themselves to really make progress conceptually?
keltan on keltan's ShortformI recently saw a tweet that asked GPT, what micro skills a person should master. I asked Claude a similar question, though had preloaded the conversation with “things a modern day Aristotle would teach a 24yo student”.
I think this is probably a valuable exercise to undertake yourself. I’m providing my prompt, and Claude’s response as an ‘appetiser’ to get you interested in this task.
—
Prompt Wonderful. Now Let's think less utilized by society. What are some more niche micro skills that you would want a person who you are tutoring to have. And this is given that you want them to be as successful as possible.
—
Claude Let me think deeply about valuable micro skills that are often overlooked but can create significant advantages in life...
Would you like me to elaborate on any of these? I find them fascinating because they're rarely taught explicitly yet can be transformative when mastered. I'm curious which ones resonate most with you?
These skills compound over time and create second-order advantages that aren't immediately obvious to others. They're also relatively timeless - they worked in ancient Athens and still work today.
jkaufman on Is the Power Grid Sustainable?you assert that you know what the monopoly spends.
First, don't we know that? It's a public company and it has to report what it spends.
But more importantly, I do generally think getting a regulated monopoly like this to become more efficient is intractable, at least in the short to medium term.
olli-jaerviniemi on Anthropic: Three Sketches of ASL-4 Safety Case ComponentsGlad to see public writing on what safety cases could look like!
I liked the section "Common precautions". Points like "wherever possible, run supervised fine-tuning on the eval", "run the eval on multiple model sizes", "ensure that we don’t train against measurements" and "all generated outputs are monitored" are commonly acknowledged as useful/important (even if obvious), so it's nice they were explicitly mentioned here.
In the AI control sketch, there's a part that I find unsatisfactory:
The situations for the game were generated using Simulation Procedure S, which we are confident generates situations from a distribution that the model cannot distinguish from the genuine deployment distribution. (See below for validating the indistinguishability of this distribution.)
It's a strong assumption that we have such a simulation procedure, at least if we are talking about General External Deployment (GED). In the Limitations section later, you write
Diversity of simulations. It seems very difficult to produce realistic simulations for a sufficiently diverse set of situations to achieve indistinguishability. Manual construction is infeasible (note that context lengths are likely to be very long). Automatically generating realistic contexts is a challenging research problem.
but I think that even this is still understating the issue: For example, during GED a user might ask a model to proofread a yet-unpublished graduate level textbook on analytic number theory she has been writing. It would be very costly to generate a diverse set of such inputs for the evaluation, and automatic generation faces the challenge of evaluation being easier than generation.
The best solution for this that I'm aware of is to train and use a trusted model to flag inputs that likely weren't in the evaluation set. Now, you mention a related technique in "[P2.5] Optional Steps."
Information hiding. Each input to an oversight model is first fed to a smaller, trusted model trained to check for clear signs that an input is either test or deployment. Such details (such as clear evidence of the date, or other side channels) are explicitly censored in all inputs such that the presence of censorship does not itself carry information about deployment status.
but I think this is far from enough: in the "textbook on analytic number theory" example, it's extremely hard to hide the fact that this input would have been costly to simulate.[1]
My answer would be to have a human oversee the model's actions on inputs flagged by the trusted model, and use such inputs to continually train and evaluate the model. But this is costly and not privacy-preserving. (The cost is acceptable if these inputs are rare enough, though. And if such inputs are common, one needs to put more effort to bridge the evaluation-deployment gap.)
I also worry about multimodality (e.g. native computer usage) making it harder to 1) make realistic simulations and 2) modify inputs in real time to hide information.
Got it, that makes sense. I think I was trying to get at something like this when I was talking about constraints/selection pressure (a system has less need to use abstractions if its compute is unconstrained or there is no selection pressure in the 'produce short/quick programs' direction) but your explanation makes this clearer. Thanks again for clearing this up!
lorxus on Some Rules for an Algebra of Bayes NetsWe’ll refer to these as “Bookkeeping Rules”, since they feel pretty minor if you’re already comfortable working with Bayes nets. Some examples:
- We can always add an arrow to a diagram (assuming it doesn’t introduce a loop), and the approximation will get no worse.
Here's something that's kept bothering me on and off for the last few months: This graphical rule immediately breaks Markov equivalence. Specifically, two DAGs are Markov-equivalent only if they share an (undirected) skeleton. (Lemma 6.1 at the link.)
If the major/only thing we care about here regarding latential Bayes nets is that our Grand Joint Distribution P[XG] factorize over (that is, satisfy) our DAG G (and all of the DAGs we can get from it by applying the rules here), then by Thm 6.2 in the link above, P is also globally/locally Markov wrt G. This holds even when P[XG]>0 is not guaranteed for some of the possible joint states in XG, unlike Hammersley-Clifford would require.
That in turn means that (Def 6.5) there's some distributions P can be such that P[XG] factors over G, but not G′=G+anonloopyextraarrow (where G′ trivially has the same vertices as G does); specifically, because G,G′ don't (quite) share a skeleton, they can't be Markov-equivalent, and because they aren't Markov-equivalent, P no longer needs to be (locally/globally) Markov wrt G′ (and in fact there must exist some P which explicitly break this), and because of that, such P need not factor over G′. Which I claim we should not want here, because (as always) we care primarily about preserving which joint probability distributions factorize over/satisfy which DAGs, and of course we probably don't get to pick whether our P is one of the ones where that break in the chain of logic matters.
shankar-sivarajan on How to cite LessWrong as an academic source?This is a little like the case of the Haruhi Problem, where a significant advance regarding the number of superpermutations was made by an anonymous poster on 4chan. In that case, the ephemerality of the post was a reasonable concern, and the solution was someone reposted the proof on ArXiv OEIS (with "Anonymous 4chan Poster" as the first author), and then cited that.
Here, you have a fixed url, so you could just follow the established conventions for citing webpages. I don't think you need any special justification for it, nor do you need to treat this as anything other than "merely another online website" (you don't think it's "snobbish or discriminatory" to pretend it's something more because you count yourself among its users?).
steve2152 on Goal: Understand IntelligenceI didn’t read it very carefully but how would you respond to the dilemma:
I’m guessing you’re in the second bullet but I’m not sure how you’re thinking about this alignment concern.