Alex G

@headinthebox You verify it separately, in layers or stages. Even with pure stuff, there are many ways to write one thing, a spectrum between "most verifiable" and "most performant". The first gives you the math of your algorithm, and you expand it when doing the more performant version.

@clayrat Absolutely, which begs the question, how do you know the function that specifies the postcondition is correct?

Perfect example of why I got disappointed with program verification. Too often the post-condition is just a pure functional version of the imperative implementation. Why bother even writing the imperative code, assuming a decent compiler will optimize the former into the latter under the covers for you.

Applying Functional Programming to make AI more reliable? HELL YEAH! AI isn’t only about big data and neural networks there’s also a world where logic leads the way... That’s where Alexander Gryzlov, proof engineer, steps in. In his talk, he’ll show how classic techniques get a fresh twist, with clever combinators doing the heavy lifting so you can focus on the ideas. From his early days as an industry developer, Alexander blends practice and theory to show how “old-school” AI and modern techniques can work together to create software that’s both reliable and a little bit magical Get to know him and his talk in depth here: lambda.world/speakers/?spea… #SymbolicAI #Agda #DependentTypes #FormalVerification

@ahh_soka There's a fairly new branch of algerbraic geometry called singular learning theory which aims to explain the theory behind deep learning, see mdlg.ai lesswrong.com/posts/fovfuFdp… sites.google.com/view/sumiowata…

how does geometry,topology etc help with DL/ML research? (and exactly who does it help? the machines in learning better or the humans in feeling smart?)

@ahh_soka @adamnemecek1 ecommons.cornell.edu/server/api/cor… numdam.org/item/ITA_2000_… cs.ioc.ee/~tarmo/papers/… arxiv.org/abs/2306.16341

@clayrat @adamnemecek1 can you point me to some specific examples?

Braid theory is really underestimated. It’s not about braids, it’s about partial commutativity, braids are a way of controlling noncommutativity.

@NickMCThree @kmett @joomy It's based on order theory, which is a "propositional" version of category theory

@kmett @joomy i might be getting too far outside my knowledge base here, but off the cuff I think the static analysis stuff like infer is in an entirely different universe of theory than algebra or category theory. Yea, the former has a lot of practical grip, i'm not convinced about the latter

Live coding Haskell and specifications! @nikivazou introduces LiquidHaskell and refinement types in this afternoon session

@dr_c0d3 it might be interesting to point out that ssreflect was introduced in its day with exactly this goal in mind, i.e., to allow easier proof maintenance by making the tactic scripts less brittle (e.g., by handling names explicitly and avoiding dependent type classes)

@clayrat Quickly skimmed through the document, this part of the conclusion confirms my views: "Proof engineering is particularly far behind software engineering with respect to maintenance, disincentivizing experts from changing the system once it has been verified."

This can't really be the future of programming and formal methods engineering, can it?

@krismicinski @miniapeur for sheaves you can try direct.mit.edu/books/oa-monog…

@dr_c0d3 have you seen arxiv.org/abs/2003.06458 ?

@clayrat I think what I am looking for is examples and analysis of how "proofs at scale" work in an organisation which is not focusing on academic research and publication.

@dr_c0d3 If there's some sort of interesting repeating pattern or higher abstraction, then ideally it should be factored out into a separate theory instead of fiddling with lists directly; like in other kinds of programming.

@dr_c0d3 Thanks! My point is that you typically don't really need to read the proof, it's essentially just shuffling around chunks of lists according to some rules; what you need to see is the definition of the theorem.

@dr_c0d3 looks like a pretty standard line of reasoning about some property holding on numerical lists under some blockchain-induced ordering, I've written dozens of proofs like that :)

@clayrat Writing, I don't doubt that, but reading?

Join us for an engaging talk by @clayrat from @IMDEA_Software as we delve into the intriguing realm of formalizing infinite programs and codata, using #Agda as our vehicle👨‍💻. Don’t miss out on this enlightening event! 📆 /cc @iLABMadrid meetup.com/fp-madrid/even…

@ariil @pandabanda у местных речевые центры очень развиты, это правда; мне тут как-то даже объясняли, что невежливо НЕ перебивать собеседника - это дескать показывает, что ты его не слушаешь

@pandabanda нет уж, прочитай! я даже специально отыскал его снова 😬

Наткнулся на совершенно новый уровень нытья новоприбывшего в Барселону IT-специалиста, которому местные своей громкой манерой разговора мешают работать из кафе.

@kreatific @ataiiam @yoheinakajima It's essentially a typelevel tree transducer automaton, only it's constructed interactively by a very big and expensive heuristic, instead of a compiler operating on high-level code.

Its funny. It "feels" similar to what it feels like looking at Assembly. Sure, if you really concentrate, you can probably figure out what's going on but then there's that "mental load/barrier" that you need to get over. I sometimes wonder whether there'll be a type of "pyramid" of apps/modules built by AI with humans being trained to describe, test, and possibly stitch pieces together to achieve an end result.

coding isn't dying, just dying as we know it once AI can build all the apps we have today easily, we'll simply start building more complex apps til they start breaking coding work will look more like a car mechanic, you're not building the car or car parts, but you know where to look when things break

@NikolayKRD @chesnolena вы просто не умеете их готовить!

@chesnolena Мне кажется у меня за всю жизнь было дай бог столько свиданий, я просто не перевариваю новых людей

@maux_de_tete какой из? :)

@clayrat первый раз понял акроним в этих твитах, горд собой!

Доделал коиндуктивный DFS и его корректность, завершил пруф "частичной" нормализации" на LH.

@vpatryshev @Signify_Tech bienvenido al sur de Europa

@Signify_Tech I would not advertise this salary level. It's a shame.

New opportunity! Scala Developer, €35,000 - €40,000 + 7k bonus - #Lisbon. tinyurl.com/2ckoeeaz