imh

Yeah, you could make the same complaints about jpeg compression.

One way to generate a sequence of heads and tails is to start with either heads or tails, then randomly choose how many of those to use before switching to the other. You're sampling the length of runs. e.g. Heads, then choosing 12113 would translate to HTTHTHHH.

It feels easier to ditch some of my biases generating a sequence this way.

What cant there be more housing development on the edge of SF...

Three of the edges of SF are water.

I was so bummed when I looked into this in college. Going for a physics degree, there was basically nowhere I could go in the spanish speaking world with classes that counted towards my degree. Which is nuts, since physics is a degree almost everywhere.

The seventy percent of New Yorkers who supported Amazon and now vent their anger also bear responsibility and must learn that the silent majority should not be silent because they can lose to the vocal minority and self-interested politicians.

This like makes me frustrated. It shouldn't be everyone's responsibility to go on twitter or wherever and be loud. Maybe politicians should put in real effort to learn about who they represent. They could engage in the same kind of polling they do during election season, rather than just paying attention to whoever is loudest.

The definition of best is irrelevant. The point is the distinction between maximizing something for an individual versus maximizing that same thing for the group.

I think your hands probably have more bacteria than your junk.

I doubt Phelps was fucking with the interviewer. I was also a competitive swimmer and my experience matches what Phelps said.

It'll reward the companies with the best groups of people, but not the best groups of companies, which I think the other comment was getting at.

I'd bet that sample efficiency is a factor in translating they most hyped bits of RL into solving IRL problems. So many business problems translate to "Learn which of these things to do, as quickly and cheaply as possible."

The best argument might be that eventually no one will trust video.

Obama talked about it this afternoon. He said "This is bad, blah blah oh no." Of course, you don't believe me because I made this up. That doesn't preclude you from believing written quotes, given the right chain of trust. It's been great to have formats like video that didn't require the chain of trust for a while, but if that time has passed, there's nothing we can do. It is hard, but in the context of text where quotes have been easy to fake for ages, we have dealt with it. It's good for everyone to be on the same page.

Thanks for the answer! Out of curiosity what about systems we can prove able to "do" basic arithmetic, ignoring whether they can talk about it?

I'm imagining the difference between a system that can show "P(n)" for any n, rather than a system that can show "P(n) for any n."

It seems like the former must come with a proof about the system. The quantifiers "for any n" have to come somewhere. If they aren't embedded within the system, do we still end up with a system that must be able to express "This sentence is not provable?"

Neat work, but doesn't behavior require colluding to be considered collusive? Otherwise this is just another neat equilibrium.

I never understood the step about how a system that can do basic arithmetic can express the "I am not provable in F" sentence. Does anyone have an ELI30 version of that?

Anything you post on Aether will be gone in about 6 months. This is nice, because no one can stalk your decade’s worth of Reddit history and figure out where you sleep.

This seems really misleading. If I can view it, I can save it. Enough people take data dumps from reddit that even if reddit decided to follow the same policy, it wouldn't mean "no one can stalk your decade’s worth of Reddit history and figure out where you sleep."

Of course I don't consider myself insulated from these platforms.

You do use them, even if you didn't sign up or agree to it, you do use their services. Your contact info is already there and every chance FB or whoever gets to learn something new about you, they take it.

That's what I mean by "Write privacy regulation." That's separate from the whole news/propaganda/content thing.

And the "first they came..." poem is a ridiculous comparison. I'm not laughing about the oppression of people who have beliefs or traits they stand up for. I'm laughing about people who don't like an unnecessary product but aren't willing to walk away.

As someone who doesn't use Facebook, Twitter, or their subsidiaries, the debate is kinda laughable. Sure break them up for antitrust. Write privacy regulation. But regulating the content shared there? If you don't like it, it's so easy to opt out, and it feels great. People write like these are utilities necessary for a good life, which sounds crazy from the outside.

Hacker news does some parts of that better, like promoting new comments higher. On reddit, a new comment is unlikely to be visible enough to get traction.

Do you have a source for the claim that NYC has more devs than silicon valley? The only source I could find was about tech jobs, not just devs. It says

[NY's] tech work force, according to most analyses, is less than half and perhaps one-third that of the Bay Area.

https://www.nytimes.com/2018/11/14/technology/new-york-tech-...

As languages with more interesting type systems come along, more powerful IDE tooling will be possible (e.g. idris 2/blodwen code completion https://twitter.com/edwinbrady/status/1050528305743622147)

More hopefully, maybe new language developers will put more work into easier and easier FFIs? How often have you thought "I wish language X could catch on, but all the useful libraries are in language Y?"

My CEO is currently out on maternity. Other people are covering. It just kinda works.

I wouldn't say that makes games orthogonal to real world problems. That's what makes them good stepping stones. Risk free "cheap" testing makes for fast research.

I totally agree about the ability to just skirt sample complexity. It's a tough one, made tougher by how early stage this work really is. We want bots to be able to match human ability and match human learning. Though they're put together, they're have very separate concerns.

For matching human ability, we're just beginning to learn techniques to get bots able to master hard tasks (e.g. incomplete information games, atari games, picking objects up). Those bots mostly learn waaaaaay slower than people. But never mastering is worse than slowly mastering, so it's early days.

On the other hand, you have people working on efficient learning. This is the question you're getting at with compute scaling arbitrarily-ish. It's more impressive if it can master a game after only playing it a small number of times. People are definitely working on this too, but for even simpler tasks. There's a lot of work right now in contextual bandits on learning fast, and that's a kind of baby-RL task. Even there, simulation tasks are super important because you really need a counterfactual to say whether you're doing well compared to alternatives.

what would that even mean?

I gave an example at the top level comment. Apply the test (blind) to known innocent samples too, and see whether it implicates them.

The problem is that the expectation of being able to transfer this stuff over is more hype driven than anything else. People hear "AI beat the best person in the world at something intellectually hard" and think that it should be smarter than people in other ways too.

That doesn't diminish work on playing games. If they had released a chess or go challenge where the board is just bigger or more pieces, that would be dumb. But this challenge is a game that is a tiny bit closer to real "problems that people care about." Solving this won't get us to problems that people care about either, but it'll get us closer. It's only an incremental step, but that's okay.

I wonder if whoever is supplying evidence via tests should be forced to (blindly) apply the tests to people other than the defendant. If your test says some random sample is more likely match than I am, it's very weak evidence.

I went to the color factory and thought it was fun without taking pictures. I also went to a gorgeous shrine on a lake near Mt Fuji and there was a line dozens of people long to take a selfie at it. The problem isn't the pop-up experiences.

Similarly, there's a ton of math in the applied side of neural network research too. It's not math vs no math. It's a question of foundational "General Theory of X" math.

I thought the idea was always more about colonization. Diversify the species extinction risk. Spread through the galaxy.

Work effort seems like a collective action problem. It's a race to the bottom, trying to beat (or not be the worst of) your team. Unions are one way to solve this (collective bargaining).

Another is just competing over workers. As the economy specializes and jobs require more and more training, I wonder if this competition is getting more limited. Industry X is treating workers better than industry Y? Too bad you can't switch because you've invested years of training (and student debt) into Y and you'd have to go back to school to get into X.

Side note: I also wonder whether that contributes to software engineers having it so good. As software eats the world, every industry needs software, so switching from the entertainment industry (netflix) to web search to social media isn't that hard.

I'm surprised to see no mention of decision paralysis. Hanging over every decision is a meta decision of how much effort to spend in making it (and so on, recursively). Given all the unpredictability the author points out, many decisions don't get much more benefit from much more thought.