gog-ma-gog

uhhhh this seems completely fucking insane?

This actually seems like a decent compromise. Sam and Greg can retain velocity on the product side without having to spin up a whole new operation in direct competition with their old levers of power, and Ilya + co can remain in possession of the keys to the kingdom.

the only thing significant figures seem to be optimizing for is debate about the right way to use significant figures

in my experience, it’s always better to eschew them entirely and simply write an explicit error margin

Articles like this on the internet are usually completely garbage, but Natalie is a really fantastic journalist. If you’re worried about woo because of the title, there isn’t any here (and quanta magazine is generally great for coverage of science)

tldr: there’s a calculation you can perform in quantum mechanics that suggests particles can tunnel through barriers faster than they could have traveled through free space at the speed of light. Scientists have now measured this more precisely, and this effect seems to hold up, but is still very small, so it’s hard to understand the ramifications of this for things like causality.

Rationalists seem extremely common amongst the $300k+/year software engineers I know. If that isn’t “successful”, then I’m not sure we’re using the same word anymore.

You can argue that the causal arrow doesn’t point in the right direction—that is, that people that are successful just happen to wear the affectation of “rationalist” because that happens to be fashionable, and not that their rationalism led them to a “successful” career, but that doesn’t seem to hold up against scrutiny, in my experience. Of those that I personally know, most have been engaged with the community since at least the golden days of lesswrong.

Scott Alexander is his pseudonym, which he freely shares, and he blogs about his psychiatry work.

I claimed no such thing :D

The number of people “qualified” to go to Harvard is probably more than 10x the current capacity of Harvard, though. Maintaining scarcity/exclusivity in the brand is more valuable than servicing the number of people technically qualified to go there, from Harvard’s perspective.

The subtlety here is that NNs do have a model, but it’s hard to see. Not just any neural network can perform as well as GPT-2–a very specific architecture can. That architecture, coupled with the data it’s trained on, implicitly represents a model, but it’s wildly obscured by the details of the architecture.

In this sense, people like Sutskever think that GPT-2 is a step on the path towards discovering the “correct” model.

It’s probably difficult to make much more progress without making extremely crisp by what you mean a “model” is, though, because I feel like it’s just as easy to move goal posts about what it means to “understand” as it does to “model”.

For example, replace every instance of “a model” in your post with “an understanding”, and it parses almost identically

To elaborate a bit: people like Marcus tend to overload/move the goal posts with what the word “understand” means. I kinda feel like in a world where we have perfectly conversational chat bots that are capable of AI complete tasks—-that if these bots look like Chinese rooms under the hood, he’ll still be complaining that they don’t “understand” anything.

I don’t think it’s unreasonable to say that if you think something that doesn’t “understand” anything can do what GPT-2 can do, then maybe your definition of “understand” doesn’t cut reality at the joints

https://nostalgebraist.tumblr.com/post/189965935059/human-ps... for an orthogonal point of view—-I feel Marcus is a bit too embroiled in this particular debate to make level-headed criticism on the merits/potential of GPT-2

iirc, no one has been able to replicate this group’s results. The “X17 particle” wiki page has a good review.

*in mice

Lorentz symmetry was already kinda-sorta baked into Maxwell’s equations. It was only a matter of time before someone took it seriously.

To my knowledge, there are still no tasks a D-Wave is better at (asymptotically) than a classical machine

It’s also a matter of signal to noise. Not very many things (other than neutrinos) are detectable when you point a detector straight down into the earth. Contrariwise, you’re swamped with signal when you point a detector at space.

...yes? Why is it not surprising?