taktoa

7 ms·

31.
by taktoa 7y ago
How are there network effects in delivery apps? It's a bipartite graph, and the number of restaurant companies is tiny compared to the number of users.
32.
by taktoa 8y ago
It could prove its own correctness, but there is still the possibility that it has a bug that causes its proof of correctness to be meaningless :-)
33.
by taktoa 8y ago
Trees are easy in Rust. It's graphs that are harder.
34.
by taktoa 8y ago
Is this not the definition of "confidence interval"? The first few Google results all define it this way...
35.
by taktoa 8y ago
I'm pretty sure the actually difficult (i.e.: not handled by the heuristics TSP solvers use in practice) examples for TSP are impractical to convert to a physical object, and I'm not sure that physics would actually do a good job
36.
by taktoa 8y ago
Stick a video camera inside the car. If an ML algorithm detects anything too far outside of the norm, send the video to a human. If the human notices damage, the riders are banned from the system. Banned rider lists could be shared between
37.
by taktoa 8y ago
The basic idea is to have an expressive programming language where all constructs are differentiable. Since the composition of diffeomorphisms is a diffeomorphism, large programs (like ray-tracers) will be differentiable as a result.
38.
by taktoa 9y ago
Not true: http://www.math.cornell.edu/~oconnor/sia.pdf https://ncatlab.org/nlab/show/synthetic+differential+geometr...
39.
by taktoa 9y ago
I'm confused by this statement. `nixpkgs` contains tens of thousands of packages, the majority of which have binary substitutions via the NixOS Hydra build farm. Usually the only time I build things from source on my NixOS laptop are w
40.
by taktoa 9y ago
If you tell me which specific piece of software you tried to package, I can probably explain what is going on. I strongly suspect that it is either not Nix's fault, or an inherent tradeoff in packaging software (i.e.: not a design flaw
41.
by taktoa 9y ago
> Then there is performance, it's very slow and uses a lot of memory making it impossible to use on a 512 MB vm, for example. No, this is just a side effect of the CLI user experience being garbage. `nix-env -i <regex>` is alm
42.
by taktoa 9y ago
Nix already does binary substitution. It just has the assumption baked in that builds don't do anything nondeterministic (and the Nix build sandbox removes a lot, but not all, of those opportunities; the ones remaining are things like
43.
by taktoa 9y ago
It "does something similar" in the sense that it doesn't obey the FHS. NixOS is otherwise a pretty radically different approach AFAIK.
44.
by taktoa 9y ago
I strongly disagree. Creating a Nix package is significantly easier than creating, e.g.: an Arch package, from my experience. packageOverrides is really nice too.
45.
by taktoa 9y ago
Why make it easy to use something bad and hard to use something good? The concept of "this variable is mutable" makes more sense than "this variable is constant" anyway; saying that a variable is mutable is like _adding
46.
by taktoa 9y ago
> Except in case of profile-guided optimization (which few people use), the compiler has one massive weakness: it doesn't know the "shape" and distribution of your data. If it did, it would be able to do quite a bit more,
47.
by taktoa 9y ago
Since when does the GPL stipulate anything about how much you can charge for GPLed software (wrt your statement about "copying fees")?
48.
by taktoa 9y ago
`cabal new-build` basically fixes the problems you have had with Cabal, and the development experience with Haskell under Nix(OS) is pretty great too.
49.
by taktoa 9y ago
See my comment above.
50.
by taktoa 9y ago
Note that Ponylang "capabilities" are actually a weak form of session types (which Graydon mentions). They have little to do with object capability theory, other than by analogy.
51.
by taktoa 9y ago
Nix(OS) is the future of package/configuration management IMO. It'd be such a shame if someone built a new OS without learning from NixOS.
52.
by taktoa 9y ago
Prolog is a programming language for turning polynomial-time algorithms into exponential ones.
53.
by taktoa 9y ago
What's "improper" about Template Haskell, other than the easily-avoided phase restriction?
54.
by taktoa 9y ago
Section 10.2 and 10.3 of this PhD thesis are pretty good: https://nixos.org/~eelco/pubs/phd-thesis.pdf
55.
by taktoa 9y ago
Nix(OS) is basically that... If only we could get upstream to start building with Nix
56.
by taktoa 9y ago
Standard Chartered has a multimillion line codebase in Haskell (technically an internal variant called Mu).
57.
by taktoa 9y ago
This is 100% false. The Calculus of Constructions is a superset of System F. You may be thinking of the fact that if you want your type system to be sound when interpreted as a logic, then termination must be guaranteed (which would break b
58.
by taktoa 10y ago
are you aware of the fact that the Δv to "fall" into the sun is _more_ than the Δv required to get to, for example, Pluto? it would be way cheaper to drop nuclear waste into Jupiter, though limited launch windows could be an issue
59.
by taktoa 10y ago
"old-school" AI is just a pretentious name for algorithms/graph theory/combinatorics, plus a bit of (now very outdated) PLT mixed in. "new-school" AI (machine learning) is just a more pretentious name for stati
60.
by taktoa 10y ago
For a while, I have interested in the idea of modifying the NixOS stdenv (standard build environment) to use a compiler that emits LLVM bitcode, and then having a function that takes any derivation to an equivalent derivation containing the

More ›