Things I've heard rationalists say

su3su2u1-deactivated20160226

Things I’ve heard rationalists say

slatestarscratchpad

I just wrote about this!

Eliezer said this a while ago, and people have been mocking him for it ever since.

su3su2u1

Then the person who said this to me was just quoting Eliezer and maybe not thinking his own thoughts. The context for this quote though, was a general denying of calories in/calories out weight loss models.

We now have a bet going where he is monitoring his calories and exercise carefully by eating soylent/mealsquares and wearing a fitbit. If I can fit an energy balance relationship to his resulting weight data, I win. If his weight loss/gain is unrelated to his energy balance, he wins.

slatestarscratchpad

I approve of this strategy!

I should add that Eliezer does not deny calorie-in calorie-out models, and as far as I know none of the serious paleo people (Taubes, etc) do either.

The position I hear from many people including Eliezer is that it’s entirely determined by how many calories you take in and out, but those are determined by appetite and activity-level-drive and these things are variable based on various factors. For example, Taubes says that eating fat makes you more satiated per calorie than eating carbs does, and that too many carbs give you insulin resistance which means that when you need energy your body will be less likely to burn fat and more likely to just make you hungrier.

I don’t like Taubes’ exact biochemical sequence of what happens, but something at least vaguely along those lines (ie there being factors that seriously affect hunger/satiety drive/metabolism) must be true to explain eg why so many medications make you gain weight as a side effect.

…
This is certainly a reference to MetaMed, which is run by a math PhD with journal access. They’re not legally allowed to “diagnose” people but they do a heck of a lot of telling doctors “Hey, have you noticed this matches all the signs and symptoms of this rare disease, maybe you’d like to test for that, hint hint?”
No, it was not a reference to MetaMed. It was someone I went to gradschool with who despite fantastic health insurance and being pretty ill refused to go the hospital, and had his wife and two of her friends reading through journal articles to try to figure out how to treat him.

That is extremely stupid and if it happens again please give them my email address ( scott[at]shireroth[dot]org ) and I will try to convince them otherwise with whatever community cred I have.

Also, I had no idea MetaMed was at all tied to Less Wrong?, and I’m not actually that familiar with their exact model or who they are really. I did look at a few reports they had put together because a client was thinking of using them to supplement info from Up-to-Date, but their report was a shorter version of the same recommendations in Up-to-Date, and they were WAY more expensive (I think the charge was per-report and not access to a report database like Up-to-Date, but it was a while ago and I’m super fuzzy on the exact details)

MetaMed was started by rationalists and is led (?) by Michael Vassar, who’s a former MIRI something-or-other-important. I worked there briefly. What you describe is definitely its main failure mode - often the patient would have a competent doctor, the literature wouldn’t have any big surprises, and then we’d give them a report that matched what everyone else said but demand lots of money for it. Having that happen once too often was one reason I didn’t really enjoy working for them as much as I should have, and I’ve kind of warned some people who aren’t super-rich away from the company for that reason. The times we did find something were fun, though.

There were also the all-too-frequent “We aren’t finding anything especially novel, but your doctor is an idiot and missed things in plain sight” issues. I complain a lot of about doctors who prescribe antipsychotics for sleep, but the things that some MetaMed customers’ doctors prescribed for sleep are infinitely worse.

More you might like

su3su2u1:

slatestarscratchpad:
A friend from medical school just emailed the following question to me, saying “You have weird friends in AI, ask them the answer to this”:
If you use a Random Forest machine learning classifier, and apply k-fold cross-validation, where do the cross-validation splits come from? Each decision tree in the random forest is already providing a 1/3 split, and the final result is just the mode of all the classifications from the trees. So where/how does the k-fold cross-validation happen?
So, weird friends in AI, anyone know the answer to this?
If it’s built in to the package you’re using then the details might vary a tiny bit, but before the random forest is made, you split the data into training and test sets. Then the trees are all built on the training set.
k-fold cross validation involves splitting the data into different training and tests sets. Say I do 100 fold cross validation, I might partition the data into 100 different sets. On the first run, I pick the first set to be my test set, and the other 99 are training. I then calculate how well the model fits that set.
Then I choose the 2nd set as validation,etc. In the end, i get 100 slightly different models with slightly different error rates on each validation set. This gives me some idea of the variance in my estimates. Does this make sense?
Edit: Also, each decision tree might not be providing a 1/3 split, depends on how you set the algorithm up. If you don’t have the computational time to cross validate, you could use the out of bag error for each tree and get a similar result.
Edit edit: I have no idea what level to aim this answer at, feel free to ask more questions.

Thanks a lot! I’ll let him know.

Source: slatestarscratchpad

A friend from medical school just emailed the following question to me, saying “You have weird friends in AI, ask them the answer to this”:

If you use a Random Forest machine learning classifier, and apply k-fold cross-validation, where do the cross-validation splits come from? Each decision tree in the random forest is already providing a 1/3 split, and the final result is just the mode of all the classifications from the trees. So where/how does the k-fold cross-validation happen?

So, weird friends in AI, anyone know the answer to this?

deusvulture

sacculetta

So there’s a Twitter bot that retweets random tweets that happen to rhyme. On election night, these slapdash poems got surprisingly poignant, if dark…

Source: sacculetta

badconlangingideas

#466

badconlangingideas

Make a conlang based on vaporwave. The script shall include full-width Latin alphabet and both Japanese and Chinese characters. Borrow random words from Japanese and Chinese languages and often alter their meaning. Many chopped up and repeated words. Numbers must always be written with at least one decimal.

slatestarscratchpad

THIS IS JUST JAPANESE.

Anonymous asked:

Is there something I can do to prevent recent random fears that seem like a very mild version of panic attacks from turning into actual panic attacks?

Consider getting a CBT workbook to better learn how to think about these.

Anonymous asked:

How can I self-treat compulsive lying? I do not want that on psychiatric records.

I’ve never treated a compulsive liar and I’m not the best person to ask, so take this only as random suggestions for things that you might want to look into.

From what I understand, psychiatry usually treats this as a symptom of something else, most often a cluster b personality disorder (eg narcissist, borderline, antisocial). These are all hard to treat, but sometimes dialectical behavioral therapy can work? Depending on whether you have any other symptoms, and whether your lying seems to be part of a wider pattern of behavior, you might want to consider either a DBT therapy group (if there’s a free one around that you can just walk into and not get a record for) or a DBT workbook (which you can buy on Amazon). Otherwise, you should try to figure out what situations make you lie and try to stay out of those situations (when possible) or plan ahead of time how you’re going to handle something that makes you want to lie (for example, if you didn’t do a task, and you know someone’s going to ask you about it and you’ll be tempted to lie, rehearse beforehand what you’re going to say, and roleplay it until it isn’t anxiety producing).

Anonymous asked:

Is there a specific time of day when Unsong updates or is it random?

Don’t worry, I’m almost done with today’s! It’s coming, I promise!

(it’s “whenever I’m in front of a computer and have finished/proofread it sufficiently”)

Anonymous asked:

I remember reading about a study where German students took a personality test, and then were told (at random) that their answers indicated either that they would faithful or unfaithful in relationships. The people who were told that they had 'unfaithful' personality types then rated faithfulness as less important than the other group. Do you know the details? What's this cognitive bias called, and has it been consistently replicated?

I think the study you’re thinking of is http://spr.sagepub.com/content/early/2013/01/23/0265407512472324.abstract . I don’t know if it has a specific name but it’s usually linked to the idea of cognitive dissonance. I don’t know if it’s been replicated.

Anonymous

Anonymous asked:

What causes people to have some values/morals and not others? Mine just seem random.

This sounds like a job for my favorite table ever!:

Anonymous

Anonymous asked:

Hello I am a rather crazy student veterinarian who recently followed you, I am the person who sent you the ask about bird lungs. I may occasionally spam you with other random facts about non-mammalian physiology and anatomy, is this ok with you???

I would be honored to hear more from a veterinarian. We owe you big time for fighting to keep this country safe.

Anonymous

worldlypositions

Santa Fe

One problem with visiting important places that are more than a few cubic meters in extent is that it is not clear what parts of the space you have to move through, doing what, to achieve successfully visiting.

You probably need to get out of the car. You probably don’t need to examine every fence post. But between those extremes, you are on your own (but for the voices of every fencepost urgently advertising the need for you to examine it, and to go on the fence post experience tour, and buy fencepost memorabilia).

And all the while you are uncomfortably wondering whether you are enjoying the Taj Mahal gift shop without realizing you should be stepping outside and seeing the Taj Mahal. Or whether you are enjoying the Taj Mahal, but would enjoy it so much more deeply, for some impossible to pin down reason, if you were just here at sunset on a Tuesday instead.

Along these lines, S suggested that a pilgrimage to Los Alamos might be more satisfying with some specific experiences one was meant to have along the way. So I tried making a scavenger hunt for our next real stop, Santa Fe, based on the internet’s views on what is important there.

Santa Fe is apparently interesting for its architecture—especially adobe and Greek revival style—its art, its food, its history. So we had to find canales and nichos and vigos, and white wood columns, brick copings, and double hung windows with divided sashes. And a turquoise stone depiction. And the state capitol building, and the oldest house in America. And something we hadn’t eaten before, and something delicious. And something we could read in Spanish. And so on.

(Hotel outside of town that would have ticked most of our adobe style boxes, and building in town that probably did.)

We probably found almost everything. We lost the piece of paper, and some of our patience for this kind of thing. I possibly learned more specific facts about architecture than in the rest of my life added together. I possibly paid more attention to my surroundings than, um, usual (..added together). I’m not sure that we found the proper experience of visiting Santa Fe. We did find a giant metal dog with a swing under it though, and who can really say which is better?Especially with the artist standing that close.

santa fe actual travel

See, that’s what the app is perfect for.