Aram Hăvărneanu
@aramh
Followers
4K
Following
17K
Media
109
Statuses
7K
Mathematical engineer bringing type safety to the cloud. Previously worked on CUE @cue_lang. JSR PC, @(R6)+; I wrote the arm64, sparc64, and Solaris Go ports.
Joined May 2009
Understanding means understanding the duality between the concrete and the abstract.
3
5
76
I think a hallmark of being new to formal proof is not realizing that you always feel like you're 30 minutes away from a QED, no matter how far you actually are. Hijacks your reward system for sure
7
10
375
Whatever potential utility might exist in LLMs doing stuff for you is more than countered by people affected by LLM delusion.
5
0
32
Turns out you can just make things up.
3
1
83
@schweingehtbree @ElliotGlazer Junk theorems can always exist as long as you have encodings. Type Theory alleviates the need for encodings in many cases, however.
1
1
11
I have to object to a few things here. First, untyped λ-calculus terms don't crash of course, but they can diverge. Since type systems for λ-calculus were primarily invented for preventing divergence, I must point this out. So it is not the case that "you will always get
One genre of nonsense I love is languages in which every possible term is well-typed. For example, in untyped λ-calculus *every* term is a valid function you can apply to *anything* else. It can't ever crash, and you can always ask stuff like `1(+, 3)` and get out an answer.
0
6
53
HDR content doesn't have any place except in a movie theatre or a similarly lighting-controlled environment where the content is the only thing you consume. On the computer it should only be enabled for full screen content, and it should be under trivial user control.
0
0
12
@ElucidationsPod @Hasen_Judi I have many problems with Docker and adjacent technologies like OCI. A Dockerfile is comprised of an ad-hoc programming language without any semantic considerations. Not only the semantics is undefined, but there has been no effort whatsoever even in trying to design useful
4
5
22
@HostOfMeta I understand how to perform focusing, what I don't understand is: 1. Why is there a need for focusing at all? Why do terms get stuck without it? Feels like a bug in the term calculus, like there are too many syntactic forms to establish a precise CH correspondence. Surely a
3
2
4
I really don't understand focusing (proof theory), but I feel like I should.
7
0
15
I wonder if ∞-categories have anything to do with the infinite-dimensional topological space I discovered for CUE's denotational semantics.
1
2
15
Type theory vs. set theory is the wrong question. I want types in my sets and sets in my types.
1
1
45
There is an even deeper truth in this. Whether you think types are sets or not perfectly correlates with whether you like dynamically-typed languages or not. Dynamic type aficionados always believe that types are sets. Even if they can't articulate exactly what types or even
In case you’re wondering what the difference is between sets and types, Thorsten Altenkirch’s got you covered.
9
5
95
I was just talking to someone today—actually minutes ago—about how IntelliJ is so much faster than VS Code even though it's written in Java then JetBrains pushed an update that completely revamped the UI, and guess what? It's dogshit slow now. Unbelievable.
20
2
215
If we had precise specifications (e.g. using dependent types) and reliable ways to prove that the code matches the specification (e.g. again using dependent type checking) then I wouldn't care about the aesthetic qualities of the code at all (e.g. whether it was produced by an
10
7
62
Particular vs. general solution is a false dichotomy. The power is having a fully general mechanism for expressing solutions that is then used to implement a very particular solution.
2
2
13
@pmddomingos @Jme_Tau If you use ℝ-valued tensors, then you can encode anything, the cardinality argument doesn't apply anymore. You can hide anything in the reals. In theory. In practice you don't use ℝ-valued tensors, you use floating point numbers with very (!) finite precision.
0
0
1
@pmddomingos @Jme_Tau If you use ℝ-valued tensors, then you can encode anything, the cardinality argument doesn't apply anymore. You can hide anything in the reals. In theory. In practice you don't use ℝ-valued tensors, you use floating point numbers with very (!) finite precision.
0
1
4
@HRenkel @Jme_Tau @pmddomingos Since Datalog itself cannot do negation without extensions, this is an insane claim. "Datalog with negation" is very different than the plain Datalog from the Tensor Logic paper.
1
0
1