LESSWRONG
LW

All of MattG2's Comments + Replies

MattG210

The diagram thing is weird, I explicitly added it back in when I posted.

Regardless, I somehow ended up posting this on the wrong account, reposted in the correct account.

Reply
MattG210

The drafting software I was using represented headings that way (as I said this wasn't written on LW). I've fixed the headings.

Reply
Recommendations vs. Guidelines
MattG250

I think this is actually a general pattern that happens in most knowledge worker careers, not only late in careers. Certainly when I was a career coach one of the key things I did to help people move up in their careers was to help them move a level up in their thinking.

I think one of the reasons that the particular meta-level up move that you're talking about happens late in careers is that at that point people who are at the top of their careers basically don't have another meta-level up they can move to understand their field - they've already made that move. So the only meta-level they can move to next is to apply the move to itself.

Reply
Recommendations vs. Guidelines
MattG250

There are people who have done this in a variety of fields but they seem to be largely niche. One reason that this may be is that most medicine literally has magic pills that allow you to apply each of the solutions, whereas most other effective process models require a deep understanding of each step to apply them.

For instance TRIZ is an attempt to create a clear process model for "How to Come up With Creative Solutions as an Engineer. " THIS is one process model for TRIZ, and each step in that process model would expand to a diagram that's... (read more)

Reply
Recommendations vs. Guidelines
MattG230
Boyd wrote about the OODA loop in his late 40's but never seemed to make the next meta level jump up to trying to instill the kind of reasoning that generated the OODA loop (or EM theory for that matter) pedagogically.

This is exactly what he did with "The Discourse on Winning and Losing"

Boyd is one of my favorite examples of a great process modeler and meta-level thinker because he did it at every level of his career:

Process modelling of why he was such a good fighter-pilot led to EM theory.

Process modelling of why EM theory worked led to O... (read more)

Reply
1romeostevensit
oh awesome I wasn't actually familiar with that. So it fits the pattern of happening later in most careers (dated 1987? he would have been 60)
Open thread, August 28 - September 3, 2017
MattG210

Is it possible to make something a terminal value? If so, how?

Reply
0RowanE
By believing it's important enough that when you come up with a system of values, you label it a terminal one. You might find that you come up with those just by analysing the values you already have and identifying some as terminal goals, but "She had long been a believer in self-perfection and self-improvement" sounds like something one decides to care about.
Philosophical Parenthood
MattG210

There seems to be a weird need in this community to over argue obvious conclusions.

This whole post seems to boil down to:

  1. You are altruistic and smart.
  2. You want more altruistic and smart people.
  3. Therefore, you should propagate your genes.

Similar to the recent "Dragon Army Baracks", which seems to boil down to:

  1. We want an effective group organization.
  2. Most effective groups seem to be hierarchical with a clear leader.
  3. Therefore, it might make sense for us to try being hierarchical with a clear leader..

I mean, I get that there's a lot of menta... (read more)

Reply
0SquirrelInHell
As for your main point, see gworley's reply, though I'm not at all opposed to making the distinction more clear. The post itself very emphatically states that this is NOT the chain of reasoning that I find compelling. In particular, its point would stand even if children of smart people were somehow exactly as smart as the population average.
3Gordon Seidoh Worley
I've thought about the meta issue you're raising before, so to respond to it directly: The trouble is most people's thinking is teleological, viz. motivated to certain ends. As such writing about an idea without addressing the teleological aspects of an idea is going to be a failure to anticipate the reader's needs and answer their questions. Thus when presenting an idea it's generally necessary to take both teleological and non-teleological approaches. To address teleology alone you need not concern yourself with substance, and to address non-teleology is to ignore your (very human) reader, thus both must be considered simultaneously. To put this another way, having a theory is literally useless if you don't know what to use it for or how to use it. Not addressing use leads to difficulty in sharing ideas, such as in academic writing in journals that has expunged all teleos and consequently fails to often engage many readers with ideas. Even more succinctly: people come for the arguments/ideas and stay for the ideas/arguments.
2pjeby
Huh. Looks like the author decided to raise the price and sell it exclusively on their own site. Kind of a pity, since it means dramatically fewer people will even know it exists. Anyway, it's Embrace The Unlovable, by Amyra Mah.
Open thread, Nov. 7 - Nov. 13, 2016
MattG200

Are you tracking your calibration with something like prediction book? You may be generally calibrated And this could have just been an instance of a low probability event happening

Reply
0MrMind
Normally I don't, but that's a good idea.
Major Life Course Change: Making Politics Less Irrational
MattG220

You're making a lot of assumptions here about what other people think.

I like Gleb's content, and think that people who criticize his methods have a point, but also at times veer away from consequentialism into virtue ethics.

Reply
1Pimgd
I agree that I am making a lot of assumptions about what other people think, but I am basing these off the voting that the community seems to apply to Gleb's posts, and the comments that he seems to receive.
Voting is like donating hundreds of thousands to charity
MattG200

So if I have a 1 in 60 million chance of being the decisive vote, and 1,000,000 other voters who also voted for the same candidate could also be seen as the "decisive vote", wouldn't that mean that my EV was 640,000/1,000,000 = .64 cents?

Intuitively it seems like 640,000 for voting is way overvalued compared to some other actions, and this diffusion of responsibility argument seems to make some sort of sense.

Reply
2siIver
Yeah, you are counting the fact that so many other people are also voting twice if you divide as described above.
4Houshalter
I don't see why you do that division. The point of being the decisive vote, is that if you didn't show up to vote, the election would have gone the other way (lets ignore ties for the moment.) You can disregard other people entirely in this model. All that matters is the expected value of your action. Which is enormous.
Open thread, Sep. 19 - Sep. 25, 2016
MattG210

You want some sort of adaptive or sequential design (right?), so the optimal design not being terribly helpful is not surprising: they're more intended for fixed up-front designing of experiments.

So after looking at the problem I'm actually working on, I realize an adaptive/sequential design isn't really what I'm after.

What I really want is a fractional factorial model that takes a prior (and minimizes regret between information learned and cumulative score). It seems like the goal of multi-armed bandit is to do exactly that, but I only want to do it ... (read more)

Reply
1gwern
I still don't understand what you're trying to do. If you're trying to maximize test scores by increasing them through picking textbooks and this is done many times, you want a multi-armed bandit to help you find what is the best textbook over the many students exposed to different combinations. If you are throwing out the information from each batch and assuming the interventions are totally different each time, then your decision is made before you do any learning and your optimal choice is simply whatever your prior says: the value of information is the subsequent decisions it affects, except you're not updating your prior so the information can't change any decisions after the first one and is worthless. Dunno. Simulation is the most general way of tackling the problem, which will work for just about anything, but can be extremely computationally expensive. There are many special cases which can reuse computations or have closed-form solutions, but must be considered on a case by case basis.
Open thread, Sep. 19 - Sep. 25, 2016
MattG270

Let's say I have a set of students, and a set of learning materials for an upcoming test. My goal is to run an experiment to see which learning materials are correlated with better scores on the test via multiple linear regression. I'm also going to make the simplifying assumption that the effects of the learning materials are independent.

I'm looking for an experimental protocol with the following conditions:

  1. I want to be able to give each student as many learning materials as possible. I don't want a simple RCT, but a factorial experiment where student

... (read more)
Reply
3gwern
You want some sort of adaptive or sequential design (right?), so the optimal design not being terribly helpful is not surprising: they're more intended for fixed up-front designing of experiments. They also tend to be oriented towards overall information or reduction of variance, which doesn't necessarily correspond to your loss function. Having priors affects the optimal design somewhat (usually, you can spend fewer datapoints on the variables with prior information; for a Bayesian experimental design, you can simulate a set of parameters from your priors and then simulate drawing n datapoints with a particular experimental design, fit the model, find your loss or your entropy/variance, record the loss/design, and repeat many times; then find the design with the best average loss.). If you are running the learning material experiment indefinitely and want to maximize cumulative test scores, then it's a multi-armed bandit and so Thompson sampling on a factorial Bayesian model will work well & handle your 3 desiderata: you set your informative priors on each learning material, model as a linear model (with interactions?), and Thompson sample from the model+data. If you want to find what set of learning materials is optimal as fast as possible by the end of your experiment, then that's the 'best-arm identification' multi-armed bandit problem. You can do a kind of Thompson sampling there too: best-arm Thompson sampling: http://imagine.enpc.fr/publications/papers/COLT10.pdf https://www.escholar.manchester.ac.uk/api/datastream?publicationPid=uk-ac-man-scw:227658&datastreamId=FULL-TEXT.PDF http://nowak.ece.wisc.edu/bestArmSurvey.pdf http://arxiv.org/pdf/1407.4443v1.pdf https://papers.nips.cc/paper/4478-multi-bandit-best-arm-identification.pdf One version goes: with the full posteriors, find the action A with the best expected loss; for all the other actions B..Z, Thompson sample their possible value; take the action with the best loss out of A..Z. This explores the othe