throwaway287391

This is good to know for London specifically.

It's not as satisfying as yours but my general technique that works everywhere in the world where cabs are competing is: ask them if they take credit cards and their reader is working before you get into the cab. If they say no you find another cab. (In my experience they never say no anyway. But I learned to do this every single time I use a cab anywhere because this scam is so unbelievably common everywhere in the world.)

Thanks for confirming that. I assumed that the downturn was supposed to be part of the shape too, by analogy with earlier(?) terms like "V-shaped recovery", but I guess not.

If I can suggest an alternative, what about "Y-shape" where you tilt your head a little bit so that the bottom of the Y points slightly to the right? (Edit: after googling this it seems I'm not the first to propose this.)

I've heard that too, but the effect doesn't seem to have been large enough to show up in any of these plots I've seen (including in the linked post). Did I miss one where it's more clear?

I swear I'm not trolling but I keep seeing this "K-shaped recovery" term and I still don't understand how I'm supposed to see a K in these plots. Am I supposed to ignore the bottom-left leg of the K? Or was COVID initially benefitting low-income people somehow?

TBH I'm not sure I'd ever heard of "describe" -- I'd always used rev-parse to get a commit ID. But now that I'm trying it, it gives me a branch/tag name rather than a commit ID (either bare "git describe" or with the other arguments mentioned above). So...not really sure?

That's kind of a misleading comparison. It takes that many characters to exactly reproduce the "hg id -i" behavior because hg chose a certain set of defaults that git didn't happen to choose. What if as an hg user I want the first 16 chars of the commit ID and don't want the "dirty +"? The hg command quickly starts to look uglier.

If you just want the current commit ID in git it's just "git rev-parse HEAD". Which is still longer and uglier than "hg id -i", yeah, but it's a fairer comparison.

Just for the record, I think "pension" in the UK means something quite different than it does in the US. A UK pension is (usually?) more like a US 401(k).

Preface: yes, absolutely, all the props in the world to this guy for doing something so thoroughly incredible and selfless and rare for the benefit of humanity. Even if I had the means, I know I'd never have the guts to do something so amazingly heroic in as extreme a way as he did.

That said (i.e. time to be the irritating pedantic little asshole, but I think it's a genuine enough question about the purpose/execution of philanthropy)... does the anonymity really warrant extra props? We know that the guy gave away essentially all of his money; does the anonymity of the individual donations really make it more selfless? If anything the headlines "Feeney gives $100M to cause X" and going on TV to talk about it etc. would likely bring more attention and donations to X, which I assume is why Bill Gates and others tend to do things that way.

Edit: ok, I've just figured out from other posts that it basically wasn't known beyond a small group of people that he was giving away most of his wealth. I didn't realize that (although it's also my first time hearing of him). I agree then, he really did spend the bulk of his life being pretty much truly anonymous and that does deserve extra props.

Pretty sure the data centers and offices are counted in the all-time carbon footprint of zero. Your second quote is talking about switching to carbon-free energy, which is different from the carbon offsets referred to by your first quote.

I had an amazing experience with 5G home broadband in the UK...while it lasted. I paid £29/month and always got the promised 300+mbps (sometimes up to 500-600) down with ~10ms ping. It was 100% rock solid and seemed to be truly unlimited, e.g. in the first month I downloaded >500GB and never noticed any throttling whatsoever.

Then they took down the tower nearest to my flat for "improvements" a couple months into the pandemic and from then on it basically didn't work at all (it was supposed to switch to 4G and it theoretically did but the best I got with that was an extremely flaky 10mbps). They originally said it'd be fixed in a week but it was completely unusable for at least a month. So now I'm back to a cable connection.

I still miss the 5G though -- there was something pretty amazing about just plugging it into the mains and suddenly having super-fast wifi with 0 installation. I personally remain optimistic about its future, unlike pretty much everyone else in this thread I guess.

Not exactly, GPT-3 uses a variant of BPE [1], so one token can correspond to a character, an entire word or more, or anything in between. The paper [2] says a token corresponds to 0.7 words on average.

[1] https://en.wikipedia.org/wiki/Byte_pair_encoding

[2] https://arxiv.org/abs/2005.14165, page 24

> difficulty level that feels just right for a kid in the early 90s.

You must've been a clever kid and/or I'm a bit dense :) When I played most of the Monkey Island games as a teenager / young adult there were definitely several times when I had to resort to GameFAQs.

It's still not really "correct" IMO, I think it needs to be "UnitIntervalValue" as someone said above. I'd expect that things of type "UnitInterval" would be intervals, not values. "UnitInterval" would be like calling the type "Int32Range" instead of "Int32".

(NB: "In real life" I would never "correct" someone for calling it "UnitInterval", it's good enough -- but given this is a pure discussion of semantics...)

I was confused because when I'm in tmux `who am i` (or `who mom hates` or `who -m`) prints nothing. I thought that might've been the joke -- I'm nobody (or something). When I detach from tmux and run it in the login shell directly it works as expected. The phrasing in the man page ("'am i' or 'mom likes' are usual.") is also pretty strange.

Are school buses not a thing anymore? The suburb I grew up in definitely didn't have a remotely decent public transit system but the school district was required (along with all public school districts in the state of Texas IIRC) to provide a school bus that stopped within a few minutes' walk of every student's home (probably with exceptions for people who actually lived outside of the school boundaries and applied to transfer in). In high school, classes started after 9am and it all seemed to work just fine.

> the university growth cap isn't just an arbitrary creation, it exists to stop more prestigious universities siphoning off a disproportionate share of students and leaving less prestigious ones in deep trouble and this seems likely to happen now.

I've never understood this UK policy personally. In the US there are no such caps and yet less-than-top-tier universities have had no trouble attracting plenty of students. Top tier US universities (e.g. Harvard), with gigantic endowments of tens of billions of dollars in some cases, could easily afford to expand their student bodies two-fold or probably even ten-fold, and yet they don't, despite having no regulations stopping them from doing so. These prestigious universities apparently feel more than enough market pressure not to expand very much, as they feel (rightly or not) that their prestige would take a hit by doing so. I don't understand what the point of creating additional regulatory pressure on top of this would be when e.g. Harvard expanding their student body could easily be a very positive life-changing event for so many students.

You can see it in action in this video: https://www.youtube.com/watch?v=So7TNRhIYJ8

Soldiers and firefighters also certainly skew less wealthy/educated/powerful, so a priori I'd say the same power dynamics apply there (though it's of course possible that the effect is smaller if you dig into the numbers).

It's also entirely possible that some people would feel driven to participate in a medical study out of a sense of duty (e.g. I'm young and healthy and want to put myself at a bit of risk to contribute to our understanding of the virus so that my much more vulnerable (grand)parents have a better chance).

> consider creating a release flag where an end user can do at least single "slow" run on a CPU on a reduced dataset, in single threaded mode, to sanity check the results and at least verify that the computational and algorithmic pipeline is sound at the most basic level.

Ok, that's a reasonable ask :) But yeah as you implied, good luck getting the average scientist, who in the best case begrudgingly uses version control, to care enough to do this.

> A lot of open sourced CS research is not reproducible.

I'm not sure if this was meant to be a counter-argument to me, but I completely agree!

> "the code still runs and gives the same output" is not the same as reproducibility.

Yes, bit-for-bit identical results are neither necessary nor sufficient for reproducibility in the usual scientific sense.

That brings up a separate issue that I didn't comment on above: the expectation that the code runs in a completely different development/execution environment (e.g. the one the reviewer is using vs. the one that the researcher used). That means making it run regardless of the OS (Windows/OSX/Linux/...) and hardware (CPU/GPU/TPU, and even within those, which one) the reviewer is using. This would be an extremely difficult if not impossible thing for even a professional software engineer to achieve. It could easily be a full time job. There are daily issues on even the most well-funded projects in machine learning by huge companies (ex: TF, PyTorch) that the latest update doesn't work on GPU X or CUDA version Y or OS Z. It's not a realistic expectation for a researcher even in computer science, let alone researchers in other fields, most of whom are already at the top of the game programming-wise if they would even think to reach for a "script" to automate repetitive data entry tasks etc.

==========

EDIT to reply to BadInformatics' reply below (I'm being rate-limited): I fully agree that a lot of ML code releases could be better about this, and it's even reasonable to expect them to do some of these more basic things like you mention. I don't agree that bit-for-bit reproducibility is a realistic standard that will get us there.

Did you read my post? I know what a seed is. Setting one is typically not enough to ensure bit-for-bit identical results in high-performance code. I gave two examples of this: CUDA GPUs (which do non-deterministic accumulation) and asynchronous threads (which won't always run operations in the same order).

Controlling randomness can be extremely difficult to get right, especially when there's anything asynchronous about the code (e.g. multiple worker threads populating a queue to load data). In machine learning, some of the most popular frameworks (e.g. TensorFlow [0]) don't offer this as a feature, and in other frameworks that do (PyTorch [1]) it will cripple the speed you get as a result as GPU accelerators rely on non-deterministic accumulation for reasonable speed.

Scientific reproducibility does not mean, and has never meant, you rerun the code and the output perfectly matches bit-for-bit every time. If you can achieve that, great -- it's certainly a useful property to have for debugging. But a much stronger and more relevant form of reproducibility for actually advancing science is running the same study e.g. on different groups of participants (or in computer science / applied math/stats / etc., with different codebases, with different model variants/hyperparameters, on different datasets) and the overall conclusions hold.

To paraphrase a comment I saw from another thread on HN: "Plenty of good science got done before modern devops came to be."

[0] https://github.com/tensorflow/tensorflow/issues/12871 https://github.com/tensorflow/tensorflow/issues/18096

[1] https://pytorch.org/docs/stable/notes/randomness.html

==========

EDIT to reply to solatic's replies below (I'm being rate-limited):

The social science arguments are probably fair (or at least I'll leave it to someone more knowledgeable to defend them if they want) -- perhaps I shouldn't have led with the example of "different groups of participants".

> If you can achieve that, for the area of study in which you conduct your experiment, it should be required. Deciding to forego formal reproducibility should be justified with a clear explanation as to why reproducibility is infeasible for your experiment, and peer-review should reject studies that could have be reproducible but weren't in practice.

This might be a reasonable thing to enforce if everyone in the field were using the same computing platform. Given that they're not (and that telling everyone that all published results have to be done using AWS with this particular machine configuration is not a tenable solution) I don't see how this could ever be a realistic requirement. Or if you don't want to enforce that the results remain identical across different platforms, what's the point of the requirement in the first place? How would it be enforced if nobody else has the exact combination of hardware/software to do so? And then even if someone does, almost inevitably there'll be some detail of the setup that the researcher didn't think to report and results will differ slightly anyway.

Besides, if you're allowing for exemptions, just about every paper in machine learning studying datasets larger than MNIST (where asynchronous prefetching of data is pretty much required to achieve decent speeds) would have a good reason to be exempt. It's possible that there are other fields where this sort of requirement would be both useful and feasible for a large amount of the research in that field, but I don't know what they are.

> Also, reading through the issues you linked points to: https://github.com/NVIDIA/framework-determinism which is a relatively recent attempt by nVidia to support deterministic computation for TensorFlow. Not perfect yet, but the effort is going there.

(From your other comment.) Yes, there exists a $300B company with an ongoing-but-incomplete funded effort of so far >6 months' work (and that's just the part they've done in public) to make one of its own APIs optionally deterministic when it's being used through a single downstream client framework. If this isn't a perfect illustration that it's not realistic to expect exact determinism from software written by individual grad students studying chemistry, I'm not sure what to say.

> I’m curious, are dedicated software assurance teams a thing in your research area?

Are these a thing in any research area? I've heard of exactly one case of an academic lab (one that was easily 99th+ percentile in terms of funding) hiring one software engineer not directly involved in leading a research effort, and when I tell other academics about this they're somewhat incredulous. (I admittedly have a bit of trouble believing it myself -- I can't imagine the incentive to work for low academic pay in an environment where you're inevitably going to feel a sense of inferiority to first year PhD students who think they're hot shit because they're doing "research".)

I haven't used FB in a while, but are there any real barriers stopping people from making fake/alt FB accounts for this purpose? If not, that seems like a pretty easy way to just not care about any of this to me. (Or was there a theory that FB wasn't already logging everything you did on an Oculus before they started making logins mandatory?)

Good points, thanks!

When does this actually matter though? Isn't growth (or shrinkage) what you'd normally really care about (e.g. this month we had 10% more DAUs than last month)? I suppose if you changed something to attract tons of new users who disproportionately use AdBlock (for example) this becomes an issue as it wouldn't show up in your metrics, but is that sort of thing common?

I suppose if nothing else it's good to know so you can immediately beef up your numbers +20% in your slide deck for VCs.

> People were trending less formal almost everywhere in the workplace, except NYC, SF and LA (or in high touch industries like real estate or law).

Sorry if this comes across as nitpicking but I'm curious what makes you think SF remains formal outside of high touch industries? AFAIK tech is the biggest industry in SF and formal clothes are far from expected there, you would not look out of place in a hoodie and shorts in an SF tech office (or even in some reasonably nice restaurants). In most tech offices you'd look like a bit of a weirdo dressed any more formal than business casual.

Sure why not? As long as no data ever leaves the device unencrypted and the encrypted data can only be decrypted by the client at the other end. Of course you'd probably have to take the app's word for it that that's actually what it's doing if you don't have the source, but that's no different from current E2E encryption offerings from WhatsApp etc.

The part I'm not sure about is whether the on-device certification that the message is "clean" couldn't be (easily) spoofed. But it would probably help curb distribution of illegal material anyway.

> The American market has proved harder to crack. The 56 Uniqlo stores in America fall far short of Yanai’s plan, in 2012, to open 200 there. They still operate at a loss.

Huh, that's surprising -- the stores I've been to are always pretty packed.

The whole "invisibility" thing throughout the article rings true for me -- I like Uniqlo (it's more or less the only place I shop for clothes now) because the quality is usually good, especially for the price, and most of the designs are nice and plain. The last thing I want my clothes to do for me is make me stand out.