hansvm

36 ms·

631.
by hansvm 1y ago
There's a certain one-to-oneness between tokens and embeddings. A token expands into a large amount of state, and processing happens on that state and nothing else. The point is that there isn't any additional state or reasoning.
632.
by hansvm 1y ago
- It eliminates one way of forging results. Without stating the seed, people could consistently fail to reproduce your results and you could always have "oops, guess I had a bad seed" as an excuse. You still have to worry about pe
633.
by hansvm 1y ago
4. No math libraries, even if their results aren't used in the deterministic path you care about (e.g., floating-point rounding mode bugs) 5. All floating-point optimizations must be hand-rolled. That's implied by (2) and "to
634.
by hansvm 1y ago
> hash is the deterministic function In most languages and libraries, hashing is still only deterministic within a given run of the program. Authors generally have no qualms "fixing" implementations over time, and some systems
635.
by hansvm 1y ago
JSON is slow, not particularly comfortable for humans to work with, uses dangerous casts by default, is especially dangerous when it crosses library or language boundaries, has the exponential escaping problem when people try to embed subme
636.
by hansvm 1y ago
"logarithmic" here refers to the number on the scale being logarithmic in the sound pressure level. Restated, power is exponential in the knob value, which roughly matches human perception of a linear increase. An actual linear fu
637.
by hansvm 1y ago
Assuming your goal is mimicking the training data, you need some mechanism for drawing from the same distribution. AR happens to provide that -- it's a particular factorization of conditional probabilities which yields the same distrib
638.
by hansvm 1y ago
Yes, as with all of these models. The only architectures which struggle with that feature are those which have a strong "distributed" aspect to their computations, where it can take much more work than programmers typically expect
639.
by hansvm 1y ago
AR doesn't inhibit long planning processes, but some popular, modern instantiations of AR have that flaw. AR in general is critical for learning the right distribution.
640.
by hansvm 1y ago
You mean the turkey voluntary extinction movement?
641.
by hansvm 1y ago
A typical city bus has something like 2500 cubic inches of tread that it burns through, compared to 650 for a Model Y. Tires typically last 500k miles, vs 50k, generously, for a Model Y. I'd said "comparable," but that was ju
642.
by hansvm 1y ago
The city bus uses tires with a harder rubber and dimensions such that the pressure at the road is less, plus its normal driving patterns have less wear than typical Tesla use. To make those sorts of calculations easy, you can ignore all the
643.
by hansvm 1y ago
C# is definitely not the only possible language, but some things stand out: 1. You can extend other people's interfaces. If you care about method chaining, _something_ like that is required (alternative solutions include syntactic supp
644.
by hansvm 1y ago
Last time I did the math, a Tesla Model Y only had 3x less tire emissions than a semi truck per distance traveled. City buses are on-par with a Tesla Model Y if you only care about mL/km tire wear.
645.
by hansvm 1y ago
Isn't that basically the history of Nevada?
646.
by hansvm 1y ago
It's O(1) if you don't need access to every permutation (common in various monte carlo applications). 64-128 bits of entropy is good enough for a lot of applications, and that's all you get from any stdlib prng, so that'
647.
by hansvm 1y ago
There's value in "being able to master that stuff," and there's value in "having mastered that stuff." The latter lets you trim a lot of possible designs from your search space nearly instantly, letting you foc
648.
by hansvm 1y ago
Not really. It's the same reason formal lectures are so much less valuable than one-on-one mentoring. An expert's value doesn't come from a bundle of facts, but from being able to figure out which facts you need to hear right
649.
by hansvm 1y ago
https://xkcd.com/1483/
650.
by hansvm 1y ago
All of that assuming they're limiting based on human-initiated requests, not the 100x requests actually generated when you click a link.
651.
by hansvm 1y ago
I was going off the posted salary ranges on their website and ignoring stuff everyone has like decent healthcare and vacation. Do they have a bonus program or stock beyond the ESPP to make up the difference?
652.
by hansvm 1y ago
Fun fact, the mucilage that makes okra slimy is very similar to the mucilage in the "marsh mallow" plant, originally used to make marshmallows. Even people who "don't like okra" usually like it pickled, if you ever
653.
by hansvm 1y ago
It looks like senior roles top out at $220k and staff at $250k, with another $10k or so in indirect financial benefits (401k matching, ESPP, etc). The pay is certainly enough to be well-off, but it's on par with what any random startup
654.
by hansvm 1y ago
"suspicion" of underage drinking or whatever then
655.
by hansvm 1y ago
Maybe a better alloy? The urban arrow also has an aluminum frame though, which is the normal recipe for snapped frames. Having broken a couple myself, and not being a racer, I always opt for steel now.
656.
by hansvm 1y ago
You can also create a function capable of doing that for any object, or one which curries the type parameter, if you'd like to reduce the boilerplate if you have many types of objects you'd like to instantiate positionally.
657.
by hansvm 1y ago
What does the LSP lack?
658.
by hansvm 1y ago
That's exactly the blog post that clicked when I put my alias method [0] together. Their other writing is delightful as well. [0] https://github.com/hmusgrave/zalias It's nothing special, just an Array-of-Str
659.
by hansvm 1y ago
This is a great post, very approachable, with excellent visualizations. We use a variation of this sort of thing at $WORK to solve a related problem, where you want to estimate some percentile from a running stream, with the constraints tha
660.
by hansvm 1y ago
This is about to give "Killed by Google" a whole new meaning.

More ›