Ed Vul

@tszzl From Dr Seuss’s less popular sequel that explored the frailties of the human condition.

amazing things will happen to you and unspeakably terrible things will happen to you and you’ll lose 99.9% of the resolution of all those moments with your mortal memory

@gershbrain smbc-comics.com/comic/be

What is the strongest argument for "emergence" in complex systems that doesn't boil down to a form of simplicity (coarse-graining is easier to think about / work with)?

@TheZvi @robertwiblin Incentive = reward gradient? So incentive gradient is 2nd derivative of reward? Useful for evaluating how treacherously steep is the way out of your local maximum.

@robertwiblin Isn't the incentive gradient the derivative of the incentives?

Reminder to apply for a job at UCSD! (social psych: apol-recruit.ucsd.edu/JPF03026, computational social science TT teaching prof: apol-recruit.ucsd.edu/JPF03004). This is our January:

Are you excited to teach computational methods to social scientists? Come join us: UCSD is hiring an open-rank teaching professor in computational social science: css.ucsd.edu/about-new/jobs… Spread the word!

Are you a Computational Social Scientist looking for a teaching position? We are hiring an open-rank (tenured/tenure-track) Teaching professor in CSS @UCSanDiego. Assistant level application: apol-recruit.ucsd.edu/JPF02583 Associate/Full level application: apol-recruit.ucsd.edu/JPF02619

🚨 New paper alert! 🚨 In work with @KelseyRAllen & Josh Tenenbaum, we study what makes people excellent tool users. We propose this requires combining physical simulation with rapid trial-and-error learning. (1/8) Website: sites.google.com/view/virtualto… Paper: pnas.org/content/117/47…

OK, so gather.town turns out to be the best part of the cogsci poster sessions! I actually "ran into" a friend while wandering between posters. Highly recommend that folks do their poster Q&A on the platform, @cogsci_soc (put poster ID on your name)! #CogSci2020

@spiantado @timothycbates @samuelmehr @GRich_Cinci @AndrewRAConway @CantlonLab Would the following be a fair summary: National heritability studies find that environment explains a low % of variance. In international comparisons there is a lot more variability in the environment, so the % attributable to the environment internationally is much higher?

@EdVul @timothycbates @samuelmehr @GRich_Cinci @AndrewRAConway @CantlonLab And this dissociation between abilities and test scores is I think highlighted in the national IQ comparisons, because they suggest that the tests probably don't mean what everyone wants them to, especially when you take them across cultures.

@spiantado @timothycbates @samuelmehr @GRich_Cinci @AndrewRAConway @CantlonLab I'm confused. There is some "general performance factor" that is quite heritable, stable over the lifetime, and predictive of life outcomes. Is the question here what to call it? Or the extent to which it is influenced by environmental factors?

@timothycbates @samuelmehr @GRich_Cinci @AndrewRAConway @CantlonLab It's like the foundational axiom of this field is the fundamental attribution error.

@anidr0id @ArvidKappas Can someone help me figure this out: Would a measurement of someone's height also fail their "invariance" criterion?

Common self-report measures in Social & Personality Psychology undergo comprehensive (N=144,496!) psychometric assessment and meh journals.sagepub.com/doi/full/10.11… (notable exception: Cacioppo's Need for Cognition) (happily not surprised @ArvidKappas)

@kausmickey @hpashler Just two doublings. 0.1% -> 15% is much further.

"The 15 per cent is not that far from the 60 per cent we need for herd immunity." It's not?

@stefanswartpet @MaxCRoser @robinhanson FWIW, here is my attempt to put an upper bound and estimate on IFR as a function of age: vulstats.ucsd.edu/IFR-italy.html

What do we know about the risk of dying from COVID-19? - Our World in Data ourworldindata.org/covid-mortalit…

@NAChristakis Are swab tests going to have sensitivity for the same duration post infection for asymptomatic individuals as symptomatic ones?

These recent data, based on exactly the kind of random, population-based testing that we should be doing & that I mentioned, from Iceland & the Netherlands, suggests the attack rate at younger ages may *not* be much lower. But story is still uncertain. twitter.com/alexandreafons… 35/

Let’s talk about the fact that both the attack rate and the death rate among the young is indeed very low with COVID19, unlike most prior pandemics. And let’s speculate about some of the biological reasons. 1/

@NAChristakis @Yale @nytopinion @DrDavidKatz Is there room for a more sustainable policy of very stringent isolation for those at high risk, while letting the young provide essential goods and services?

Is Our Fight Against Coronavirus Worse Than the Disease? nyti.ms/2J1v53h Excellent outline of hard trade-offs we face via @Yale @nytopinion @DrDavidKatz. It’s not just the trade-off between health and economics. A devastated economy itself has bad health effects.

@JeffRouder @JuliaHaaf Maybe a strategy for a proof. Consider the rank order permutations: yyxx, yxyx, xyyx, xyxy, yxxy, xxyy. Each equally likely, and each sum(x>y) {2, 1, 0} is represented equally frequently in the permutations. yielding a uniform distribution.

@gershbrain mp3 compression. Similar idea using CSF for video here: ieeexplore.ieee.org/document/77323… Volume sliders respect Weber's law and are logarithmic.

Can anyone point me to some good references on applications of psychophysics (e.g., Weber's law) to the design of real-world telecommunication systems?

@mcxfrank It's the original Rumsfeld uncertainty: en.wikipedia.org/wiki/Knightian…

@EdVul I don’t think I know that term - is that different than Rumsfeld uncertainty?

Hey Bayesian folks: How do you think about out-of-sample validation? With my scientist hat on, I want to know how my model does OOS - freeze parameters and evaluate. But with my Bayesian hat on, my best estimates are given by *combining* training+validation and fitting jointly...

@mcxfrank Yes, that sounds exactly right. Although put this way, it sounds like I have to reconcile my instinct here with my aversion to other invocations of Knightian uncertainty.

@EdVul It’s like “known unknowns” vs “unknown unknowns”?

@mcxfrank I think it's the same situation regardless of how many levels. You are still specifying *some* hypothesis space, and your estimates are restricted to that hypothesis space. That hypothesis space may be wrong.

@EdVul Yeah that is intuitive. But level up one level and make it a model selection problem. Then shouldn’t you just pool data across the two datasets for model selection?