DanielFilan

Wiki Contributions

Comments

Privacy and Manipulation

It's plausible that literal priests and therapists have useful knowledge about this that you could find out by googling and talking to acquaintances, respectively.

Even if you're right, you're wrong

Indeed! The post would be boring if none of the bullet points were legit.

Even if you're right, you're wrong

You're not wrong. This post wasn't really meant literally.

Speaking in the language of the post:

Well, look. Let's put to the side whether or not sin^2x + cos^2x is actually 1 or not. In today's culture, the obvious and natural interpretation of what you said is that sin^2x - cos^2x = 0. But that's a damaging belief for the future engineers of America to have, that could seriously harm their faith in the math education they received at our upstanding public schools, and so it's irresponsible for you to go around saying "sin^2x + cos^2x = 1" without clarifying exactly what you do or don't mean.

The Meta-Puzzle

That was my solution.

The Meta-Puzzle

I didn't think of that one!

The Meta-Puzzle

I actually thought about including that: similarly, in American logic, you can't take an arbitrary claim P and come up with a sentence S such that you can derive P from "S is true" and also from "S is false".

The Meta-Puzzle

Depending on how the 'always lie' part is defined, a liar could say something 'impossible' like 'I am neither single nor married'.

I mean that everything they say is false.

both the honest and the liars will say 'I speak the truth'.

And they also both say "I worship God".

Everything Studies on Cynical Theories

Regarding the authors' attempts to get papers published in these journals, the review doesn't make it seem like the book relies on that experiment being valid (and the review itself does not) - it just talks about various features of these fields and theorizes about their causes and effects. I also don't think that their experiment was 'bad science' in the sense of being uninformative. If 'grievance studies' journals are willing to publish bad papers, that does tell you something about those journals, even if 'hard science' journals are also willing to publish bad papers (which we know, thanks to the replication crisis and bloggers like Andrew Gelman, that they are). Also, Wikipedia says "By the time of the reveal, 4 of their 20 papers had been published; 3 had been accepted but not yet published; 6 had been rejected; and 7 were still under review.". It seems unfair to include the papers that were under review in the denominator, since their efforts ended early, so I'd evaluate their success rate at 1 in 2, rather than 1 in 3, which isn't so bad.

The line of arguments that critical studies are illiberal because they remove the focus from the individuals, their personal choices and responsibilities to structural and systemic tendencies seems to prove too much.

This is really not what the review portrays the arguments to be, so I'm having difficulty engaging with this paragraph. Could you perhaps quote an example of that argument in the book or in the review that you think is invalid?

But a more important point is that in this methaphor wokeness and social justice are our immune system against fashism.

I'd say that liberalism is a sufficient immune system - altho I'm obviously interested in ways that it isn't.

I don't see how creating a rallying flag for liberals to stand against social justice in a culture war is going to help. On the contrary, this leads to evaporation of group beliefs and more radicalisation of the left due to toxoplasma of rage.

I think the idea is to give a certain strain of thinking a name, and analyse what its like, to make it easier for people to figure out if that strand of thinking is somehow bad and avoid it if it is. Presumably you're sometimes in favour of this kind of thing, so I'd like to know what you think makes this effort different

I don't think you can make a case against default ideology in science from the liberal position.

I think liberalism allows the idea that many people can have a wrong worldview!

Regarding how the 'second secularism' deals with issues like de facto segregation in the US: I agree that that's the sort of thing that critical social justice cares about, but it's also something that liberalism can discuss and grapple with. As you mention, in order to understand the problem you probably also need to understand racism, but that doesn't automatically mean that things other than critical social justice can't deal with the problem, or that critical social justice frames are going to be successful (e.g. it might make you think that people only bring up house values as a pretext for racism, when it seems pretty intuitive to me that people actually do care about how much money they have).

A final note: I get the sense that you maybe think I wrote this review. I actually didn't, but I mostly liked it, somewhat mooting the point.

What exactly is GPT-3's base objective?

Expected return in a particular environment/distribution? Or not? If not, then you may be in a deployment context where you aren't updating the weights anymore and so there is no expected return

I think you might be misunderstanding this? My take is that "return" is just the discounted sum of future rewards, which you can (in an idealized setting) think of as a mathematical function of the future trajectory of the system. So it's still well-defined even when you aren't updating weights.

What exactly is GPT-3's base objective?

I continue to think that the Risks from Learned Optimization terminology is really good, for the specific case that it's talking about. The problem is just that it's not general enough to handle all possible ways of training a model using machine learning.

GPT-3 was trained using supervised learning, which I would have thought was a pretty standard way of training a model using machine learning. What training scenarios do you think the Risks from Learned Optimization terminology can handle, and what's the difference between those and the way GPT-3 was trained?

Load More