chrchang523

I’m actually not criticizing SBF here.

I am criticizing people, possibly including myself, who should have developed more realistic and convincing models clarifying why his behavior was increasingly suboptimal despite his near-linear personal utility function.

His behavior doesn’t really look like self-deception to me. It looks like blindness to poorly articulated limitations of a mathematical model he was taught and understands very well, and I suspect he wouldn’t have been so blind without the “poorly” above.

In SBF’s case, I think there was actually an academic failure: there was something very important that he should have been taught, but wasn’t.

https://mobile.twitter.com/SBF_FTX/status/133725070407583334...

https://twitter.com/SBF_FTX/status/1299553547596431360

These are excellent topics for classroom discussion. What I believe should have followed is a stronger justification for using something like the Kelly criterion despite a personal near-linear utility function, on the basis that e.g. not going all the way down to zero asymmetrically preserves access to future investment opportunities, and similar grounds. I think plenty of people intuitively understand this.

But instead, nobody was able to explain to him why he couldn’t throw out the book on risk management, and he acted accordingly.

This is incidentally a valid reason for HN to blacklist the NYT domain, at least until a formal reversal of the policy.

I finally started seriously using a M1 work laptop yesterday, and I'm impressed. More than twice as fast on a compute-intensive job as my personal 2015 MBP, with a binary compiled for x86 and with hand-coded SIMD instructions.

That was my previous experience, but have you tried linking to AMD AOCL recently? I would not expect the performance gap between Intel MKL and AMD AOCL to still be as large as you describe on a Zen 3.

I’ve actually heard that bullying has been dramatically decreased in public schools over the last generation. (People complain about feminization of education, and I suspect they have a valid point, but that trend comes with some upsides.)

I use a double-ended stack allocator for most allocations in personal C projects. The second end adds a significant amount of flexibility; temporary allocations can accumulate at one end of the stack without preventing a longer-lasting allocation from being made later (yet before the temporary allocations can be freed) on the other end.

"An example is the Smith-Waterman algorithm for genomics processing"

Sure, they can advertise that.

But are the vector units on Intel/AMD/ARM processors really not good enough?

"Debunking the 100X GPU vs. CPU Myth" is a somewhat controversial paper, but it lines up with my experience with common genomics workflows.

This model is not consistent with what we've seen in Japan over the past several decades.

Yes, when I learned about this practice, that significantly changed my understanding of the NIMBY problem's solution space.

Although high property taxes incidentally constrain NIMBY behavior, they're a cure worse than the disease. It is appropriate for a city like Toronto to be proud of being able to operate with low property taxes. (I'm not going to defend the way CA Proposition 13 distributes the property tax burden between longtime residents and newcomers; but low property taxes on longtime residents should not be considered a problem, ceteris paribus.)

This is a bit overstated. Yes, FPGA accelerators can be used effectively in some common genomics workflows. However, my experience is that the right software running on regular Intel/AMD/ARM processors is very competitive with FPGA-using solutions.

My understanding is that the Winograd minimal filtering algorithms used in cuDNN are different from the O(n^2.37) Coppersmith-Winograd-descended matrix multiplication algorithms. But I acknowledge that these can be considered cousins, produced by the same line of research.

Sometimes. In the case of matrix multiplication, there is a pretty large backlog of "galactic algorithms" going down to ~O(n^2.373) that haven't yet led to real-world improvements. Strassen's ~O(n^2.807) algorithm is only considered over the basic O(n^3) strategy for n>1000.

Same here. My use case was simple — a bunch of static pages — so render.com isn’t even charging me.

For the same reason, you’d never intentionally stop at $0 (unless the deck is exhausted). You can always get back to $0 by drawing the rest of the deck, and there’s a 50% chance of going up a dollar on the next card.

The problem, at least in California, is that there is too much of an incentive for a long-time resident to not sell even if they want to move out permanently. Because they are paying extremely low property taxes that cannot be passed on to the next buyer, so even with high overhead, the efficient outcome is still for them to rent their place out instead of selling. So existing houses go on the market at a very low rate -- often only when the owners die.

Combine this with the number of Bay Area potential buyers who are relatively insensitive to interest rates due to being able to make very large down payments, and of course the lack of new SFH construction relative to SFH demand... I don't think Bay Area SFH prices are actually going to drop much.

I'm more optimistic about affordability of denser housing. I'm unsure how many Americans are actually happy with that, though (which is part of why it's relatively affordable...).

This works, but I'd add the following: always keep your eyes open for "dirty work" that you seem to enjoy more than most other people. That represents an opportunity to make disproportionate contributions.

Relatedly, you should prioritize building good working relationships with others who specialize in different, complementary types of "dirty work".

In your case, how much compatibility is lost from using C99 over C89? (I'm comfortable with C99 and still write some low-level multiplatform code in it, but losing the ability to declare variables where they're first used is a bridge too far for me...)

One has to be particularly careful with climate policy because fuel prices have a long track record of high political relevance.

(2009)

See also: https://www.lesswrong.com/posts/wkuDgmpxwbu2M2k3w/you-have-a...

With no weirdness points, you duplicate others' work. With too many weirdness points, you're almost certainly wrong. The sweet spot for innovation is in between.

That isn't the issue here; the requested "Taiwan, China"/"Chinese Taipei" labeling does still distinguish the source.

(2019)

It sounds like you're misevaluating your technical interview performance. You can feel like you're "bombing" an interview, while still performing above expectation. Some interviewers ask hard technical questions, don't expect most candidates to complete them within the allotted time, and give "hire" recommendations for some (but of course not all) that don't.

Regarding the bit “compress” operation, this is supported in hardware (PEXT instruction) by Haswell and newer processors, though unfortunately most AMD processors have poor implementations. I’ve found it to be quite handy when implementing subsetting operations.

That's a contributor to how the math team continues to be successful today, but the more interesting question is how they initially became world-class ~15 years ago. Frazer is clearly responsible for the latter.

At a time when only 4% of the population had graduate degrees, if 25% of married grad-degree-holders were married to another grad-degree-holder, that would constitute a large preference in favor of other graduates.

“Something like 7%” is noticeably less than twice 4%, so no, I didn’t make that assumption. (I also explicitly stated that this was a back-of-the-napkin-quality analysis.)

The math is trickier than that:

- If 4% of parents have degrees, something like 7% of the next generation has at least one parent with a degree. Because children have two parents, and only one of those two parents needs to have a degree. So the correct number to compare 20% against is ~7%, not 4%.

- Suppose a child of a degree-holder is 3 times as likely to become an economics PhD as a child of no degree-holders. Then, if 7% of children have a degree-holding parent, you would expect (7 * 3) / (7 * 3 + 93) = 21 / 114 = ~18.4% of economics PhDs to have a degree-holding parent, not 21%. If 25% of children have a degree-holding parent, you'd expect (25 * 3) / (25 * 3 + 75) = 75 / 150 = 50% of economics PhDs to have a degree-holding parent, not 75%.

Eyeballing the last few numbers, it does look like the value of having a degree-holding parent may have increased significantly, from a ~3.3x multiplier to almost 6x. But the error bars on this back-of-the-napkin analysis are pretty wide.

I believe Lemire's point is that the compiler is still able to inline the function despite the generic definition, so at least in this somewhat narrow case generics are a zero-cost abstraction.