haeffin

They themselves said it was trained on NVIDIA chips, so I’m not sure where you got that it wasn’t. It was trained on the less capable versions sold for the Chinese market.

The contents changed between then and now.

Why do you think there's a longer interval between inspections for Teslas? The interval for Hauptuntersuchung in Germany (which this article is about) is 2 years (except for the first one which happens after 3 years), regardless of car brand.

And a windshield wiper needing replacement wouldn't count as a "significant fault" that they force you to come back for before getting certified.

Unfortunately, some companies' HR departments actively encourage "trivia questions", because they ask interviewers to ask every candidate the same questions so that a "fair" comparison between candidates.

Why wouldn't it have good coordination? A bot has access to a perfect model of how the other bot would act - itself.

Also, computer engines didn't seal the doom of human players in chess and in go, so I don't get why it would do so in dota.

It's not a pro game, those are still amateur players (some of them ex-pros).

Physics and other disciplines were broadly using arxiv before ML did it. When I switched from physics to ML (CS) I was surprised that arxiv isn't used more.

Depends on what your industry is. If you're compiling c++ for mobile, clang is pretty much your only choice for iOS and the default choice for Android ...

> Because our training system Rapid is very general, we were able to teach OpenAI Five many complex skills since June simply by integrating new features and randomizations. Many people pointed out that wards and Roshan were particularly important to include — and now we’ve done so. We’ve also increased the hero pool to 18 heroes. Many commenters thought these improvements would take another year.

The linked commenters thought that getting to "real dota" (more than 100 heroes, captains mode instead of random, ...) would take another year. So I don't think it's fair to make that statement.

Edit: Don't get me wrong, I think the improvements are very nice, but pointing to people saying "these people thought we would need a year, we did it in under a month!" is not what you should do if you didn't actually do what the linked people stated.

Because it's absurd to throw all EU countries into one pot in this regard, especially if you include (and focus on) the UK?

It was mainly sold as "prevent illegal foreign gambling sites from making a profit without paying back".

Geometry Dash, still probably my favorite game on the iPad, switches between left-to-right and right-to-left suddenly in the middle of levels and it's messing with my brain.

I'm not saying "Google's Awesome!". My point is that the view you presented is far more biased than what I get from my colleagues and friends that stayed in academia. People running around like headless chickens in the "MS buys GitHub" thread was unfounded, and it is unfounded here.

Supporting independent academic research like https://ai.googleblog.com/2018/03/google-faculty-research-aw...?

Why do you think Google doesn't encourage publishing papers? Google has a very big amount of papers they publish ... as an example, https://ai.googleblog.com/2017/12/google-at-nips-2017.html

Usually when people speak about open source, they include not only not being able to view the source, but also the possibility to modify, distribute and use it (and modified versions) in the requirements for a license to be labeled an open source license.

That's not a straight line (it only looks like that on this map).

A car that stops for no reason is a safety risk for the cars that follow it, so that's not a good option either.

When did you do your PhD? When we collaborated for papers we used Subversion (at the start of my PhD) & git (towards the end) both for code and papers (in latex).

I get dizzy about half an hour into using the HoloLens. I am not sure why, I think it may be because the tracking/rendering doesn't quite keep up with fast head motions making the holograms "inconsistent" with the real world. It might also be that it's because of the accomodation/convergence mismatch ...

Which is why there is the "And more importantly you can revoke it (at any point) and the site can't deny or degrade the service (unless the data is strictly necessary for a specific action related to the service)." point - you're not allowed to deny access to a newspaper article if somebody does not consent.

It does already and did for a long time, but customs are a hassle.

Nice - the "Detectron operators currently do not have CPU implementation; a GPU system is required." on https://github.com/facebookresearch/Detectron/blob/master/IN... is outdated then? Or is it the "nearly all" that's behind this statement?

That's lots of names of companies and no examples of "best technology building best practices".

That would be awesome. But - is a tutorial enough? From what I gather from the readme the custom ops have cuda implementations and no CPU ones, which you'd like to have as a fallback on mobile (even if metal/opengl implementations exist), or am I wrong?

caffe2 (which this is built on) was switched from bsd+patents to Apache2 a while ago too.

It's a bit disappointing, when caffe2 was released it was stated that mobile is a big focus, but things like this don't support mobile (even though some of this was demoed by FB on a phone).

Where did you get that? The STL implementations I know definitely don't. See https://github.com/llvm-mirror/libcxx/blob/master/include/li... for an example - __list_node_base contains the pointers, __list_node contains the object and derives from __list_node_base, so allocating the node with object and pointers is one allocation.

Or maybe it is because it isn't actually mainly opencv code and the top comment didn't actually look at it but nobody checked and just upvoted.

It varies a lot by country and field (if they do classes or not). I didn't have to do classes when I did my PhD.