Calavar

You were probably thinking about this piece on nondeterminism in attention by Thinking Machines: https://thinkingmachines.ai/blog/defeating-nondeterminism-in...

Const is the reverse.

Constness is infectious down the stack (the callee of a const function must be const) while asyncness is infectious up the stack (the caller of an async function must be async). So you can gradually add constness to subsections of a codebase while refactoring, only touching those local parts of the codebase. As opposed to async, where adding a single call to an async function requires you to touch all functions back up to main

Of course it's useful, that's why function modifiers like 'const' or 'virtual' (thinking from a C++ perspective) are widely seen as useful, but making one function virtual doesn't force you to propagate that all the way up the call tree.

If you are serving web pages, you are soliciting GET requests, kind of like ordering a package is soliciting a delivery.

"Taking" versus "giving" is neither here nor there for this discussion. The question is are you expressing a preference on etiquette versus a hard rule that must be followed. I personally believe robots.txt is the former, and I say that as someone who serves more pages than they scrape

I agree. It always surprises me when people are indignant about scrapers ignoring robots.txt and throw around words like "theft" and "abuse."

robots.txt is a polite request to please not scrape these pages because it's probably not going to be productive. It was never meant to be a binding agreement, otherwise there would be a stricter protocol around it.

It's kind of like leaving a note for the deliveryman saying please don't leave packages on the porch. It's fine for low stakes situations, but if package security is of utmost importance to you, you should arrange to get it certified or to pick it up at the delivery center. Likewise if enforcing a rule of no scraping is of utmost importance you need to require an API token or some other form of authentication before you serve the pages.

There are many more ways to degrade model performance than to enhance it, so I would expect the vast majority of bugs to lead to artificially reduced accuracy, not artificially increased accuracy.

So if PyTorch is full of numerical flaws, that would likely mean many models with mediocre/borderline performance were discarded (never published) because they just failed to meet the threshold where the authors felt it was worth their time to package it up for a mid-tier conference. A finding that many would-be mediocre papers are actually slightly less mediocre than believed would be an utterly unremarkable conclusion and I believe that's why we haven't seen a bombshell analysis of PyTorch flaws and reproducibility at NeurIPS.

A software error in, say, a stats routine or a data preprocessing routine would be a different story because the degrees of freedom are fewer, leaving a greater probability of an error hitting a path that pushes a result to look artificially better as opposed to artificially worse

Likely bad because the conviction will still show up in background checks. (A pardon doesn't entail expungement.) As you said, this won't be an issue for CZ because he's not going to be looking for the kinds of positions where HR screens the resumes

Not if you're always getting the tranfusion from the same person

Apparently yes: https://abcnews.go.com/International/canadian-mans-lottery-w...

That's my understanding. Uber takes 25%, but by default that's offset by increasing the on-app price 25% relative to the in-store price, and the owner has to explicitly opt out of that behavior. So at the end of the day, they should be getting the same amount as in store orders unless they opted out of the markup, right?

Could you explain further? The restaurant gets the order at the same price, Uber adds delivery fees and a 25% markup on the cost of food. Why would these numbers not work out in the kitchen's favor? Maybe I'm missing something. Genuinely curious.

America itself and the opportunity to get your foot in the door with a student visa is the prestige factor for many students - the prestige of the university is secondary. That allows a lit of third tier universities to fill their master's programs with international students

He does say that around 5:32 in the video. He says his driver told him there were two cars on the loop that day, and the other car wasn't in service because it was being used for training.

The Rails development stack wasn't nearly as convoluted in 2015. A lot of this stuff leaked over from the JS ecosystem over the years

I think the difference is PG faked it until he could make it work at some future endpoint, Charlie Javice gave up on trying to make it work and instead faked it at the endpoint

Yes, there is certaintly data to make that argument. The argument that inflammation is a better predictor of heart disease than cholesterol is a reasonable one. On the other hand the argument that cholesterol is not a good predictor of heart disease and cholesterol lowering therapy is a fraud is not reasonable in the face of the evidence.

That's a hot take of a blog post.

Extraordinary claims require extraordinary evidence. The cholesterol to heart disease link is one of the best attested in medical science [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]

And yet when making this very extraordinary claim, the author fails to cite any quantitative data for or against. He does not even attempt to build a qualitative argument by proposing a mechanistic theory of why cholesterol is unlikely to be causative of heart disease. Then he goes on to claim that doctors don't have hard evidence to show that statins reduce the incidence of heart disease, despite the fact that such evidence exists [5]. The post is just 10 paragraphs of fluff that boil down to 'don't trust the medico-industrial complex'

Honestly, I think that blog post is a litmus test on scientific literacy - What convinces you more, data and numbers and charts and tests of statistical significance, or rail-against-the-machine rhetoric and a few scary sounding quotes provided without the associated context?

[1] https://jamanetwork.com/journals/jama/article-abstract/19216...

[2] https://pubmed.ncbi.nlm.nih.gov/25815993/

[3] https://jamanetwork.com/journals/jamacardiology/article-abst...

[4] https://www.ahajournals.org/doi/abs/10.1161/01.CIR.67.4.730

[5] https://jamanetwork.com/journals/jama/fullarticle/2678614

[6] https://pubmed.ncbi.nlm.nih.gov/32507339/

[7] https://www.tandfonline.com/doi/abs/10.1080/07315724.2008.10...

[8] https://pubmed.ncbi.nlm.nih.gov/18061058/

[9] https://www.jacc.org/doi/abs/10.1016/j.jacc.2022.03.384

[10] https://www.nature.com/articles/s41598-021-00020-3

[11] https://www.ahajournals.org/doi/full/10.1161/JAHA.123.030496

[12] https://www.nature.com/articles/s41467-024-46686-x

[13] https://www.sciencedirect.com/science/article/pii/S002191502...

[14] https://link.springer.com/article/10.1186/s12872-021-01971-1

[15] https://www.jstage.jst.go.jp/article/jat/31/3/31_64369/_arti...

The Oregon Health Care Study followed patients for 2 years initially, then it was expanded out to 3 years. That's an absurdly short interval.

The idea is that increased primary care services will have benefits 10 or 15 years down the line by preventing chronic disease from reaching a critical state. For example, preventing prediabetes from reaching diabetes and then diabetic end stage renal disease (which would require dialysis at a cost of 5 figures per person per month). You're not going to see that over 2 to 3 years.

The Gulf States are seeing absolutely excellent ROI for the 1b they pooled together for a 747.

Yep, QUIC is what I was thinking about when I wrote my additional comment, but there are many examples, as others have pointed out.

> Chrome is an excellent browser with leading standards support.

Google learned it can be "standards compliant" if it submits a draft spec to WHATWG/W3C, and while the comment and revision process is still ongoing, roll out those features in Chrome and start using them in YouTube, Gmail, Google docs, and AMP. Now Firefox and Safari are forced to implement those draft specs as well or users will leave in droves because Google websites are broken. Soon enough, Google's draft spec is standardized with minimal revisions because it's already out there in the wild.

The debate, revision, and multistakeholder aspects of the standards process have been effectively bypassed, a la IE6 and ActiveX, but Chrome can claim to be on the cutting edge of standards compliance. This is a case of Goodharts's law.

If you get into options trading, make sure to have a hard stopping point.

> Hence Leetcode is a reasonably good test for the job. If it didn't actually work, it would've been discarded by companies long ago.

I see it differently. I wouldn't say it's reasonably good, I'd say it's a terrible metric that's very tenuously correlated with on the job success, but most of the other metrics for evaluating fresh grads are even worse. In the land of the blind the one eyed man is king.

> If you can show that you can become excellent at one thing, there's a good chance you can become excellent at another thing.

Eh. As someone who did tech and then medicine, a lot great doctors would make terrible software engineers and vice versa. Some things, like work ethic and organization, are going to increase your odds of success at nearly any task, but there's plenty other skills that are not nearly as transferable. For example, being good at memorizing long lists of obscure facts is a great skill for a doctor, not so much for a software engineer. Strong spatial reasoning is helpful for a software developer specializing in algorithms, but largely useless for, say, an oncologist.

I am a physician, so I can say this with a high degree of confidence.

That snippet is referring to the circle of Willis*, which is a "backup" circuit that can route around a blockage to the blood flow to the brain on one side.

The thing is the circle of Willis is tiny and near vestigial (there is a substantial fraction of the population where it doesn't even make a full circuit), whereas the carotid is one of the largest blood vessels in the body. The circle of Willis isn't nearly large enough to reroute all that flow. It has to be made bigger over time through a process called collateralization, and that's a process that happens over months to years, not minutes or seconds.

In short, the circle of Willis will save you from years of high cholesterol that lead to a huge cholesterol plaque completely blocking off one of your carotids. It won't save you from your carotid being severed by a bullet.

*And some other tinier vessels, but mainly the circle of Willis

It's not a case of losing blood, it's a case of failing to move blood to the right place. If the shot took out the carotid, then (nearly) 50% of the blood supply to his brain is gone because of a piping failure. That can absolutely cause instantaneous loss of consciousness, no direct brain trauma necessary.

This is very different than bleeding from, say, a major artery in a leg. In that case the issue isn't loss of piping to the brain, it's losing blood until the total blood volume in the body isn't sufficient to maintain a workable blood pressure, and yes that can take multiple minutes before a person loses consciousness.

I don't buy that. You can't have it both ways. You can't discount PHP4/5 criticism because PHP7+ is good and then in the same breath hold PHP above other tech stacks for passing the test of time, not when PHP7 is younger than Rails, Django, and React (and even Vue). Those frameworks have passed the test of time more than PHP7+.

I've heard that a lot, but my gripe with modern PHP is what's the hook? What feature does it offer that other scripting languages don't?

- Can you use the same rendering logic on client and server like JavaScript?

- Does it have the performance of Java or C# or Go?

- Does it allow for expressive DSLs like Ruby?

- Does it have an extensive, cutting edge ML ecosystem like Python?

Back in 2010, PHP's stand out features were very straightforward installation, good support for templating, and a large collection of C library wrappers (even if the interfaces were a bit raw). Those hooks were compelling enough to make people put up with the PHP4 core language.

But other languages have caught up on those three points, which leaves modern PHP utterly unremarkable. It doesn't have any feature that stands out enough to make me want to switch back.

> It's important to understand that only a small percentage of academics will ever get tenure. The rest will keep toiling away on increasingly poorly paid and desperate postdocs until they finally age out and decide to take a job in industry.

There's also a good chunk of people who fail to advance past the assistant professor level, which is pre-tenure at US institutions (not sure about other countries). And it's up or out, so if you're an assistant professor and you don't get tenure within a certain number of years, you lose your job.

Version control took a while because most early version control systems were brittle and had poor developer UX. Once we got mercurial and git and nice web UIs, the transition was actually pretty fast IMHO.

The same could be true for coding agents too, or maybe not. Time will tell.

China is the birthplace of the term 996. Of course it's not the birthplace of people being coerced into unhealthy work hours - that's been around for thousands of years.