tel

13 ms·

91.
by tel 3y ago
Ko is very common in Go, though. You can't validly repeat the board position (simplified way of handling Ko rules) but if tree search failed to evaluate ko positions correctly it'd be easy to construct situations where the AI woul
92.
by tel 3y ago
There are a few differences, depending on how you evolve from the CBPV core, but basically other types may be eager or lazy. Functions can be eager (and thus statically know to represent an eager call tree) and sum types can be lazy (deferr
93.
by tel 3y ago
The variance component, implied volatility, is more often than not treated as the _output_ of the equation. By looking at the prices of options you can determine what the market currently, implicitly, estimates the future variance of underl
94.
by tel 3y ago
I recently finished my first "in anger" project with an Observable notebook. This involved learning Observable Plot, Arquero, relearning bits of Javascript, and integrating it with a Rust based simulator that's my data genera
95.
by tel 3y ago
I think that’s basically correct, the FFT representation is a better feature representation.
96.
by tel 3y ago
I don't know if it's a generalized result, but the Circuits team at Anthropic has a very compelling thesis: the first phase of descent corresponds to the model memorizing data points, the second phase corresponds to it shifting ge
97.
by tel 3y ago
I haven't dug into this much at all, but I am aware that some teams have been looking for methods to either discover or train in symmetries and invariances into latent spaces. For instance, in an image recognition model one might expec
98.
by tel 3y ago
It's been a while since I was steeped enough in this to answer comprehensively, but per memory, up to performance concerns, they're equivalent in inductive settings. Haskell ends up being a bad place to talk about this (or a great
99.
by tel 3y ago
Just to play an argument out: Your future living situation may be in part dependent on the economic strength of the nation you're living in, and thus there's collective onus on ensuring that future. It's reasonable to believe
100.
by tel 3y ago
Agreed, though being an expert witness is its own skill. You need to be able to coordinate and/or write reports which will hold up under scrutiny in court and also be willing and able to stand for depositions and testimony. This also m
101.
by tel 3y ago
I think that's untrue and unfair. I don't think anyone quite knows what attention is so completely as to simplify it to "just a kernel smoothing". For a great example, the Transformer Circuits team have 2022 research sho
102.
by tel 3y ago
I don't think we disagree, and I do think what I argue is sufficient for image generation models to fail to render hands well. What you add---that they are very complex---is true, I believe, but I avoided using it to argue as I'm
103.
by tel 3y ago
I think it's because hands have the dual properties of being extremely important to us and not primarily visual. Our hands are our primary mechanism for interacting with the world in a conscious and directed fashion. We devote a lot of
104.
by tel 3y ago
That definition of abs has merit. In some spaces we are able first to define only an “inner product” between elements p(a, b) and then follow on by naming the length of an element to be sqrt(p(a, a)). One trick about that inner product is t
105.
by tel 3y ago
Part of the issue is that there is no simple Why. To invent one is to either petition history or invent some perspective. It's not a bad idea to do these things, didactically, but they're unnecessary for the material. Or, said ano
106.
by tel 3y ago
Immediate giveaway is that the fireball has a lazy eye. Then you can look at the keyboard.
107.
by tel 3y ago
Eh, sure it is. Until it isn't. I'm not using this libraries, but I'm glad they exist.
108.
by tel 3y ago
I would definitely be in favor of an even faster `SmallBump` variant which assumes that the allocations are small w.r.t. usize::MAX for some additional speed. I also wouldn't mind the a default "fast" bump allocator library t
109.
by tel 3y ago
Yes I am aware of async drop proposals. And the point is not to handle a single value being dropped but to facilitate invariants during an abrupt tear down. Today, when I am writing a task which needs tear down I need to hand it a way to si
110.
by tel 3y ago
Now you lose determinism in tear down though.
111.
by tel 3y ago
Honestly, this is me too. The only thing I’d like to also see is OTP-like supervisors and Trio-like nurseries. They each have their use and they’re totally user land concerns.
112.
by tel 3y ago
Re: existential quantification and async function types It'd be very nice to be able to use `impl` in more locations, representing a type which needs not be known to the user but is constant. This is a common occurrence and may let us
113.
by tel 3y ago
Async traits come to mind immediately, generally needing more capability to existentially quantify Future types without penalty. Async function types are a mess to write out. More control over heap allocations in async/await futures (w
114.
by tel 3y ago
I think Rust’s async stuff is a little half baked now but I have hope that it will be improved as time goes on. In the mean time it is a little annoying to use, but I don’t mind designing against it by default. I feel less architecturally c
115.
by tel 3y ago
That works for channels, but being able to wait other asynchronous things is better. Timeouts for instance. We could imagine extending this to arbitrary poll-able things. And now we have futures, kind of.
116.
by tel 3y ago
Waiting asynchronously on multiple channels/signals. Heterogenous select is really nice.
117.
by tel 3y ago
It's a useful way of measuring how different two probability distributions are. If F and G are distributions, Kl(F||G) is non-negative real number which is larger if F and G are less similar. In a lot of statistical estimation procedur
118.
by tel 3y ago
“Superposition” seems to be another term used for this. Although, it implies a mechanism instead of just a behavior.
119.
by tel 3y ago
> why did no one come up with this before? Because the author is intimately familiar with the softmax function from work outside of ML, and plausibly nobody who’s investigating these issues is remotely as familiar I doubt that is true. S
120.
by tel 3y ago
People use both, for reasons of tradition, ergonomics, practicality, available toolset. Boldface is obviously common in places where BB is unavailable, more restrictive, more difficult. Web publishing is a great example. BB is common, in my

More ›