jlebar

I was 100% prepared to believe that the length of empty string in js is 7.

Then upon opening the post I was 100% ready to believe that js has three different string length functions that all handle Unicode differently.

For example, to evaluate something as simple as 1.3+2.4, you need an exact model of your target architecture's fp unit.

If you want to cross compile, generate code for x86 while running the compiler on anything other than x86, this means you need to implement 80-bit long double fp in software.

If you want to cross compile for PowerPC, you get to build software support for their bespoke "double double" format, which is not IEEE754 128 bit but a totally different thing.

Some GPUs (and also CPUs) flush fp denormals, so you also have to handle this in your constant folding.

And all that is just to handle addition, one small part of constant folding.

Prior art: https://www.arthurganson.com/concrete-1

Waymo | Bay Area and Remote within the USA | Full-time

I'm hiring for the team I lead at Waymo. We're GPU experts who care about latency on the car.

Before coming to Waymo, I lead the CPU/GPU parts of XLA, the compiler for TensorFlow that also powers JAX. We're using XLA and other technologies to make ML and non-ML GPU code run fast on the car.

Ideally we're looking for someone who's comfortable hacking on high-ish level TensorFlow and low-level XLA (or similar frameworks, like TVM), to help with e.g. speeding up int8 stuff. But I'm open to considering strong candidates without this specific experience -- all of this stuff is learnable.

Apply here https://waymo.com/joinus/4764940/ and maybe mention that you saw this on HN! Also if you have questions I'll try to check this thread and reply.

I've found it really useful for explaining mathematical concepts using the notation and terminology I'm comfortable with.

For example, GPT-4 gave me the first explanation of backwards-input and backwards-filter convolutions that I've been able to understand. This was because I was able to start it off by explaining how I understand forward convolutions, and it explained the bw convs in the same way.

Astonishingly good.

> Should language be precise, or nice? [...] Maybe freedom of thought, and the ugliness beyond, is something that should only be allowed in ones own head?

I don't think that asking people to say "died by suicide" rather than "committed suicide" is a question of precision vs niceness. Both are plenty precise, we agree they describe exactly the same IRL event.

I also don't think that asking people to say "died by suicide" rather than "committed suicide" is an affront to freedom of thought, for the same reason.

That said, sure, yes, my parents always told me to "think before you speak" and sometimes "you can think it but don't say it". We all think things that we shouldn't say, usually because saying those things will hurt someone. If you insist on saying something after being informed that it hurts someone (e.g., not using their pronouns, or referring to a painful moment in their life [e.g. the suicide of a loved one] in a way that hurts them) then, yeah, that's kind of the thing my parents tried to teach me not to do.

Right. And I am saying, Deaf people, and other people who don't think they have a disability, are asking for us not to say that they do when they don't think they do. And the answer from some quarters seems to be, I think "objectivity" is on my side, screw your conception of your identity.

It's not that different from refusing to use someone's pronouns. I think that both are a mean way to respond to a fellow human.

The entire point of this linguistic exercise is that your opinion that you would not give up your ability to hear is no more or less valid than someone's opinion to the contrary.

> But.. disability is abnormal and negative? How tortured of a definition of 'normal' must one use to make blindness normal? And I suspect if the author of that sentence were offered, say, $100,000 to undergo a painless but irreversible blinding surgery, they would very quickly admit blindness is a negative

As you can verify with some Googling, there are many Deaf people who choose not to undergo surgery to gain the ability to hear. (This isn't a universal preference by any means, but it's also not fringe.)

> The mental contortions required to conclude that "committed suicide" trivializes something, but "died by suicide" doesn't, are so extreme that I cannot believe anyone honestly holds that opinion.

I think the notion is that "committed suicide" is like "committed a crime" and harkens back to the Christian notion that suicide is a crime (against God). And so it's moralizing language that blames the person for killing themselves.

You can verify that this is a common notion among people who work in the field by Googling. For example, the first hit when I search for "why shouldn't we say committed suicide" is https://www.irmi.com/articles/expert-commentary/language-mat... which says:

> For example, the phrase "committed suicide" is frowned on because it harks back to an era when suicide was considered a sin or a crime. Think about the times when we use the word "commit": "commit adultery" or "commit murder."

> That leaves as the most likely conclusion that this document is not an honest effort to eliminate 'harm', but a mine-field, a list of sins and prohibitions, to be used as ammo against opponents.

It's interesting that you arrived at this conclusion before doing a minimal amount of research!

> What is so hard about the concept of allowing people to be wrong about things?

Because democratic society requires us to agree on basic facts in order to function.

"People are wrong about" things like:

- who got more votes in the last US presidential election

- whether future elections in the US will be fair

- whether global warming is real

- whether Democratic politicians are literal lizard people who molest children

We have seen that when we can't agree on these basic facts, people take actions like

- attempting to overthrow the seat of government

- attempted assassination of political figures (or their spouses)

- election of politicians who say that they will overthrow elections that don't go in favor of their party

- enacting policies that threaten to create a climate catastrophe.

The reason that Musk's tweet at Hillary repeating the conspiracy theory that Paul Pelosi's assailant was a male prostitute is a problem is because, in providing an alternative (false) explanation for the attack, it contributes to hiding the truth on this issue. The truth being that we can draw a straight line from politicians' words to this attack.

And since we vote for those politicians, if people are wrong about this, then yeah, it does matter.

I will never forget the conversation I had with someone about competitive Excel.

Them: It's a real thing, I was watching on YouTube!

Me: I know, it's pretty awesome/hilarious.

Them: Yeah, and they treat it like a real e-sport!

Me: With announcers and everything!

Them: I know. "Helloooo sheeeetheads!"

Me: .......yeah that's a parody.

> A while back I invested in 3M because it's a company that makes so many amazing and industry leading products and I thought they'd do well. A couple years later and the stock is just lagging.

The relevant question isn't whether they make great products, but whether their stock is undervalued.

We in the tech world are used to "great products" -> "stock price always goes up", but in the "real world" there's no reason for that to be true. "Stock price goes up" should, from an efficient-markets PoV, reflect new information.

I remember being disappointed by this book.

I wanted to read about algorithms for choosing a random float in [0,1], and IIRC its only suggestion was randint()/INT_MAX. This is wrong in part because the smallest nonzero value you'll ever choose is 1/INT_MAX, which is much larger than the smallest nonzero float.

Is there a better book out there? I was hoping for something fabulous like Hacker's Delight but for floating point.

If you like that you may also like https://graphics.stanford.edu/~seander/bithacks.html

(Also "Hackers Delight" if you really can't get enough of it.)

Presumably they're talking about putative electric heating, not cooking.

> What most people don’t know is: Gas fees on Ethereum are supposed to be high.

Count me among those not in the know!

I...really don't get it.

Bitcoin was supposed to be a currency. Then we realized that it's too slow and too energy-hungry for that. So we retcon'ed it into, it's a store of value!

Etherium was supposed to be a place for smart contracts. Then we realized it's too slow and gas-hungry for that. So...we retcon'ed it, too?

I do "love" how a wrong, one-sentence comment spawns so much argument.

So many person-hours spent arguing this point online. So much of our lives. I'm as guilty as anyone.

Yes, "someone's wrong on the Internet", haha. But really. This guy isn't arguing in good faith. Frankly at this point nobody arguing online that climate change is not real is arguing in good faith.

Can we find a way to channel our collective energy towards productive climate action? Serious question.

> Why is no one bothering to claim it?

I think of that old joke:

>>>>

The young economist looks down and sees a $20 bill on the street and says, "Hey, look a twenty-dollar bill!"

Without even looking, his older and wiser colleague replies, "Nonsense. If there had been a twenty-dollar lying on the street, someone would have already picked it up by now."

>>>>>

Surely there is not a $1M bill lying on the ground, ripe for the taking. That's why I'm not bothering to claim it.

Death is not the only risk from this disease.

Young people are at risk of getting long covid, for example. https://www.mattbell.us/delta-and-long-covid/

Yeah, sorry, I misunderstood.

Do you think it's impossible that the virus actually did infect you the second time, just asymptomatically?

> There is no scientific proof, e.g. RCT, that lockdowns work

Nor that parachutes work!

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC300808/

> Who cares how many people are vaccinated, what the goal should be is reducing hospitalizations, deaths and long-Covid

I mean, sure, but in our toolbox of interventions, vaccination is far and away the best one we have in service of those goals.

Right?

If you lose friends, family, and love interests because of your position on covid vaccines, I hate to break it to you, but this consequence was not imposed by a corporation. It was imposed by your friends, family, and love interests, who, yes, probably find your stance hurtful.

You don't have a right to a relationship with anyone.

> if you're under 50 and have no comorbidity, you're not at risk.

This is exactly the kind of false information the YouTube ban is about. You're proving how we need it.

Do you take seriously the proposition that more than 100k Americans could die from the vaccine without us noticing? This would easily put it in the top 10 causes of death in America. And presumably this w would be happening worldwide, at an even greater scale.

More than 2 billion people have gotten a vaccine worldwide.

Extraordinary claims require extraordinary evidence. VAERS is self-reported and (as this shows) not reliable data.

Honestly I think you're proving the need for this ban on YouTube. You seem to be trying to get it right, but you've been hoodwinked by the misinformation.

It is indeed hard to get it right, and this is killing people.

People like you who had a mild case may benefit most from the vaccine, because you may be the least likely to have developed long-lasting antibodies.

You've seen what the virus can do. Why chance it?

I'm sorry to hear about your wife. I hope she gets better.

https://thehill.com/opinion/healthcare/574284-natural-covid-...

> they will have to start practice serious self-censorship or get banned from FAANG-land, isolating them from friends, family and potential love interests.

You mean, there could be consequences for our words and actions?

Sure, absolutely. The point the author is making is that since you're more likely to get an antibody response with the vaccine, you're getting a benefit from the vaccine that there's a decent chance you won't get from catching the virus.

> How about referencing scientific sources rather than opinion pieces from political news sites.

The dude is a MD PhD who's been on NPR, NYT, etc. He's literally a scientist who is a source -- a scientific source.

But sure, we can talk journal articles.

Where is your "less than 2%" number coming from? I see this in the first Nature article:

> Of the 125 subjects exposed to SARS-CoV-2 according to the baseline ground truth definition, 101 (80.8%) participated to the May serosurvey. Among them, 93.5% (86 out of 92, 95% CI 86.3–97.6%), 84.2% (85 out of 101, 95% CI 75.5–90.7%), and 100% (92 out of 92, 95% CI 96.1–100%) had a positive result for Abbott, DiaSorin and Roche, respectively, whereas 44.9% (44 out of 98, 95% CI 34.8–55.3%) had a neutralising titre greater than 1:40 (1/dil). In November, 86 subjects (68.8%) were tested again, all of them except one (98.8%) tested positive to at least one serological assay.

They're saying that 44.9% of people infected had neutralizing antibodies at the level they recognize, right? They re-tested the same people six months later, and >98% of them tested positive for one assay. This doesn't speak to the 98% number you cite above. Maybe you're looking elsewhere.

I'm not going to be sealioned into going through studies if you can't point me to the section that supports your point.

Anyway for the article's 1/3 number, he also cites a study by the CDC, a fact which I guess you've chosen to ignore? https://wwwnc.cdc.gov/eid/article/27/9/21-1042_article