Veedrac

87 ms·

961.
by Veedrac 8y ago
Your entire argument seems to be "it's impossible to do this because we don't know how to do it." I don't get how people argue about the plausibility of future par-human AIs and say things like "we don't e
962.
by Veedrac 8y ago
One thing I would love personally is the ability to restrict JS usage to a CPU limit, which increases in response to certain events and gradually reduces to 0 if unused. This could be * A number of seconds on page load, * a fraction of a se
963.
by Veedrac 8y ago
The issue is not just that it's ‘slow’, but that it's frequently slower even than std::map. My suspicion is that the issue is its use of power-of-2 sizes, which is a terrible idea for binary search because it causes cache line ali
964.
by Veedrac 8y ago
You don't need ‘more’ tiers per se, because WASM can be fully compiled from the outset. Two tiers should be fine.
965.
by Veedrac 8y ago
WASM is a poor target for hardware, for reasons that don't matter at all to JITs. It has an unbounded stack, its branch instructions use tags rather than relative offsets, its calls use a function table which would be doubly-indirect i
966.
by Veedrac 8y ago
It makes sense since WASM's SIMD isn't ready yet, and a paper saying “WASM is slower because SIMD isn't ready yet” would be rather uninformative to browser vendors.
967.
by Veedrac 8y ago
Summary: > Root Cause Analysis and Advice for Implementers: > We conduct a forensic analysis with the aid of performance counter results to identify the root causes of this performance gap. We find the following results: (1) code comp
968.
by Veedrac 8y ago
Incredibly, this was an even dumber post than I expected. For its length it's hard to pinpoint any wrong argument made, because AFAICT it's mostly just a semicoherent stream of unjustified assertions and associations, but it does
969.
by Veedrac 8y ago
Competitive Pokemon with teambuilding would be something fresh I don't think existing techniques have figured out yet. Not super prestigious, but IMO still a good intermediate goal.
970.
by Veedrac 8y ago
> For a big, announced presentation, they should be at a stage to stream live games (like OpenAI did with Dota). This is such an inane complaint. They didn't do it all live because they have results they want to show and don't
971.
by Veedrac 8y ago
https://youtu.be/cUTMhmVh1qs?t=3218
972.
by Veedrac 8y ago
I thought they said 350ms. E: https://youtu.be/cUTMhmVh1qs?t=3272
973.
by Veedrac 8y ago
http://zero.sjeng.org/ for AlphaZero playing Go, and http://lczero.org/ is at least par with AlphaZero Chess based off match reproductions.
974.
by Veedrac 8y ago
Quick summary of the talk: A specialist number representation is made for exact representation of values in geometric calculations (think CAD). Numbers are represented as sums of rational multiples of cos(iπ/2n). Exact summation, multi
975.
by Veedrac 8y ago
I'm not sure, but I suspect your best alternative to 32-bit floats/floating-bar-rationals or 64-bit rationals when you have no FPU is a 64-bit fixed-point value with compressed floating-point representations if you need to store i
976.
by Veedrac 8y ago
It's very rare not to need GCD reduction, though; multiplying values burns through bits extremely fast. Only doing factor-2 reduction isn't infeasible but you'd be further damaging an already iffy type.
977.
Are Sunk Costs Fallacies?(gwern.net)
3 points by Veedrac 8y ago|0 comments
978.
by Veedrac 8y ago
Floating point is also built on integer math, and cheaper operations at that.
979.
by Veedrac 8y ago
The publication mentioned: https://www.federalregister.gov/documents/2018/11/19/2018-25... Relevant excerpt: The representative general categories of technology for which Commerce currently seeks to dete
980.
by Veedrac 8y ago
The pop effect can be tamed with, for example, font-display: optional;, which allows the first page load to use the fallback fonts.
981.
by Veedrac 8y ago
First, this isn't designed as a programming font. It's designed as a UI font. Second, they have a toggle for that, it's on the linked page.
982.
by Veedrac 8y ago
If you're putting floats in a dict, you can't use NaNs anyway, since they aren't equal to each other.
983.
by Veedrac 8y ago
> I immedieatly thought it was because float64 couldn't cover that number so I told him to use longdouble dtype (it's 80 bits in Intel I think) and the problem solved. This is a very dangerous habit. If 64 bits isn't enoug
984.
by Veedrac 8y ago
Plugging my own explanation because HN seemed to like it: https://news.ycombinator.com/item?id=15360485
985.
by Veedrac 8y ago
I suspected you meant lossless, but no, Titanfall literally did include 35GB of uncompressed audio. Their reason was to lower the CPU overhead from decompression, which sounds like nonsense to me.
986.
by Veedrac 8y ago
FWIW, overly long iterator comprehensions, or unwieldy chains of them, are common in Python, but a local generator function is much cleaner most of the time in my experience. The difference is slight here but the rule of thumb never goes wr
987.
by Veedrac 8y ago
I did, anySatisfy (\s -> p (False : s)) ==> (\s -> p (False : s)) (find (\s -> p (False : s))) ==> p (False : find (\s -> p (False : s)))
988.
by Veedrac 8y ago
I found it easier to read when `anySatisfy` was inlined: find :: (BitSequence -> Bool) -> BitSequence find p = if p (False : find (\s -> p (False : s))) then False : find (\s -> p (False : s)) el
989.
by Veedrac 8y ago
Then you'll love https://stackoverflow.com/questions/37361145/deoptimizing-a-...
990.
by Veedrac 8y ago
Python uses/used eval for namedtuple at least in part for performance.

More ›