chrchang523

Joined 9 years ago

Karma

591

Submissions

255

Member since

Mar 8

View on Hacker News

Recent submissions

Comment · 2 days ago

(2015)

Comment · 1 month ago

This is, of course, known as the Anthropic Principle…

Comment · 5 months ago

and you CAN use static_cast to convert from void*; this silently keeps working if you refactor the void* into a matching-type pointer later, while raising a compilation error if you refactor to a different-type pointer.

Comment · 6 months ago

Other things worth noting about denormal numbers:

- It’s not just ‘old’ FPUs that handle them verrry slowly. Benchmark this aspect of your target processors if this detail matters.

- Most modern processors provide a “flush to zero” denormal handling mode (and sometimes you can separately specify “denormals are zero”, relevant when e.g. loading old data). However, math libraries usually haven’t been written with this mode in mind, so you need to be careful with this.

Comment · 1 year ago

Note that BGZF solves gzip’s speed problem (libdeflate + parallel compression/decompression) without breaking compatibility, and usually the hit to compression ratio is tolerable.

Comment · 1 year ago

Yes, though I think tooling could be better; if I had more spare time I'd write a linter which flagged defers in loops that didn't come with an accompanying comment.

Comment · 2 years ago

The problem with cgo is the high function-call overhead; you only want to use it for fairly big chunks of work. Calling an assembly function from Go is a lot cheaper.

https://pkg.go.dev/github.com/grailbio/base/simd has some work I’ve done in this vein.

Comment · 2 years ago

I found it useful to walk through evaluation of a few elementary instances of this class using simpler methods, to put the main result in perspective. Specifically, replace the initial 3 exponent with 0 or 1.

If the exponent is 0, then you have the sum 1/2 + 1/4 + 1/8 + 1/16 + 1/32 + ..., from Zeno's most famous paradox (https://en.wikipedia.org/wiki/Zeno%27s_paradoxes ). If you are fortunate, you previously learned that this converges to 1, and played around with this enough in your head to have a solid understanding of why. If you are less fortunate, I recommend pausing to digest this result.

Then, if the exponent is 1, you have the sum 1/2 + 2/4 + 3/8 + 4/16 + 5/32 + ... .

What happens if we subtract (1/2 + 1/4 + 1/8 + 1/16 + 1/32 + ...) from it? We have (1/4 + 2/8 + 3/16 + 4/32 + ...) left over.

Then, if we subtract (1/4 + 1/8 + 1/16 + 1/32 + ...) from the latter, we still have (1/8 + 2/16 + 3/32 + ...) left over.

Then, if we subtract (1/8 + 1/16 + 1/32 + ...) from the latter, we still have (1/16 + 2/32 + ...) left over.

Continuing in this fashion, we end up subtracting off

(1/2 + 1/4 + 1/8 + 1/16 + 1/32 + ...) + (1/4 + 1/8 + 1/16 + 1/32 + ...) + (1/8 + 1/16 + 1/32 + ...) + (1/16 + 1/32 + ...) + (1/32 + ...) + ...

and this converges to the main sum. And, from the exponent-0 result, we know this is just 1 + 1/2 + 1/4 + 1/8 + 1/16 + ...

Comment · 2 years ago

Two-bit values are common in bioinformatics, and I’ve found the ability to efficiently convert between packed arrays of 1- and 2-bit values to be valuable in that domain.

Comment · 2 years ago

One question: it is possible for the XOR of two consecutive floating-point numbers to have 32-63 leading zeros; the numbers 32-63 do not fit in 5 bits. I imagine this is treated by Gorilla like 31 leading zeros?

Comment · 2 years ago

(2008)

Comment · 2 years ago

The mod 1000 is actually a consequence of the test format: all answers are integers in [0, 999], you fill in 3 digit-bubbles.

Comment · 3 years ago

Suppose the value of a network to an individual user is proportional to the number of users. Then the total value of the network, summed across all its users, is proportional to the square of the number of users.

Comment · 3 years ago

Yup, and that's why I'd consider this a "101" essay. The larger exponential-growth trends (e.g. Moore's Law) practically always have a microstructure with many sigmoid curves. After you've encountered your first exponential, the "201" lesson about saturation becomes important.

Comment · 3 years ago

The slave-owners were slave-owners because they had military control over the slaves. (Most of the time, anyway; some slave revolts were successful.)

It is not clear how long humans will retain control over AIs.