lalaithion

10 ms·

91.
by lalaithion 3y ago
Ignoring the fact that this comment has a completely different bar than OP, I’m not asking about chips, I’m asking about desktops. Is there a desktop for sale at any price, from any manufacturer or custom builder, that goes 2 TB ram?
92.
by lalaithion 3y ago
Example of a desktop computer that comes with 32 TB ram?
93.
by lalaithion 3y ago
The author mentions prql as an alternative to sql in the post. I personally find prql much more fluent than sql.
94.
by lalaithion 3y ago
The things that still confuse me about transformers is: 1. Why do we _add_ the positional embedding to the semantic embedding? It seems like it means certain semantic directions are irreversibly with certain positions. 2. I don't under
95.
by lalaithion 3y ago
It would be nice to have performance numbers. Right now the only thing I can find on performance is "Cedar is fast and scalable. The policy structure allows your policies to be indexed for quick retrieval. The design also supports fast
96.
by lalaithion 3y ago
I mean, the same is true with the C runtime.
97.
by lalaithion 3y ago
Has anyone tried the Scott Alexander AI bet prompts? 1. A stained glass picture of a woman in a library with a raven on her shoulder with a key in its mouth 2. An oil painting of a man in a factory looking at a cat wearing a top hat 3. A di
98.
by lalaithion 4y ago
NRVO is a completely transparent optimization that Rust tries to do whenever possible. I don’t know what sort of case you can get into with object identity but I suspect it’s obviated by making moved-out-of values untouchable at compile tim
99.
by lalaithion 4y ago
Deeply nested objects grow linearly with the size of history, not exponentially.
100.
by lalaithion 4y ago
And the way you'd do it with this library is // very complex and powerful API effing_mad::effects! { Read { fn read_to_string(bug: &mut String) -> Result<usize>; } }
101.
by lalaithion 4y ago
Here's an example from outside the tokio library: https://github.com/google/tarpc/issues/297
102.
by lalaithion 4y ago
It’s not a big deal because the ecosystem has coalesced around tokio, but it’s coalesced around tokio partially because it’s a big deal.
103.
by lalaithion 4y ago
This is what I find most annoying about rust's async ecosystem; many functions depend on being called within a certain executor, but this is not documented in the type at all. You should not provide, and arguably not be able to provide
104.
by lalaithion 4y ago
The “plain text” spec is in a weird encoding, casing mojibake when I view it in my browser.
105.
by lalaithion 4y ago
If they had homes, it wouldn't be called homelessness. You can take fentanyl in an apartment.
106.
by lalaithion 4y ago
Yeah, I mean thats an option, but I basically expect that to cause a y2k-like level of emergency refactorizations as everyone realizes that their software isn't built for it. On the other hand, we already have leap seconds. They'r
107.
by lalaithion 4y ago
Fair. It would be more accurate to say that people want the time of midnight and noon to be consistent across ~decades and centuries (and if the leap second isn't going to happen the leap minute or hour almost certainly won't...).
108.
by lalaithion 4y ago
We have a monotonically increasing time, TAI (International Atomic Time). The problem is that people want 00:00 each day to correspond with midnight, and so we need either (a) varying seconds or (b) leap seconds. So we have UTC (Coordinated
109.
by lalaithion 4y ago
Pretty hard. The key to this implementation is Haskell's laziness, so you have to implement laziness in Javascript, which will probably not be very ergonomic. You can write let loeb = (structure => { let inner = () =>
110.
by lalaithion 4y ago
Public domain software would be considered “commercial” under the DoD’s definition.
111.
by lalaithion 4y ago
The key is that Rust lets you define safe interfaces, so you can separate your data races and union juggling, which need to be unsafe, from your business logic, which basically never needs to be unsafe.
112.
by lalaithion 4y ago
It doesn't have to be dynamic dispatch; you could also do static dispatch at compile time based on type information.
113.
by lalaithion 4y ago
If you have two functions abracadabra(foo) and foo.abracadabra() which do different things , you should rename one of those functions tbh.
114.
by lalaithion 4y ago
There's also the media encoder hardware accelerator, which isn't quite `gemm`, but certainly contains hardware that performs `mm`s.
115.
by lalaithion 4y ago
The key of the idea is you're not asserting that predicates on values are true, you're recognizing whether a value (of a type with few guarantees) conforms to the guarantees of a type (implicitly, a type with more guarantees), and
116.
by lalaithion 4y ago
> In particular, if the model is trained on multiple news stories about the same event, it has no way of knowing that these texts all describe the same thing, and it cannot differentiate it from several texts describing similar but unrel
117.
by lalaithion 4y ago
Really? About 30% of the bugs we find in our entirely-Golang codebase are something that is trivially solved by a more advanced type system, and it was even higher back when I worked on Python.
118.
by lalaithion 4y ago
Haskell could have (I don't know a library that implements this, but it's ~trivial) do x <- open filename `errorContext` ("failed to open " <> filename) y <- readBytes x `errorContext` &qu
119.
by lalaithion 4y ago
Yes, that is the solution; I'm complaining that it wasn't even mentioned as a possibility in the blog post.
120.
by lalaithion 4y ago
The problem with this is that _you've lost all the wrap messages_! In a good golang codebase, you get human-readable stack-trace with the relevant variables printed inline on every error. It's super useful and every "Result&q

More ›