Comments by "Vitaly L" (@vitalyl1327) on "Continuous Delivery" channel.

@jamesgood7894 a lot of cases when people were unable to express a simple idea even in a pseudocode (anyway, no reason to ask about any specific languages, this is not an important knowledge). Is it that much to ask? And I'm not talking about dumb and pointless intevriew questions like the infamous FizzBuzz or asking to implement a double-linked list, no. Just any kind of a problem with more than two steps in its solution.
1
@vincentvogelaar6015 LLMs do have reasoning abilities. Just give them a Prolog, and let them write, test and debug proofs.
1
@vincentvogelaar6015 apparently you do not understand how to use LLMs. They're not any different from our own minds - we cannot reason either, unless we use tools, such as formal logic. So give LLMs their tools - give them an ability to write down reasoning step by step, to verify the reasoning using formal methods (as in - make them write down the steps as HoL proofs or Prolog predicates). I'm using LLMs for solving engineering problems, and reasoning is a crucial part of it. Even very small models (like Phi-3) are perfectly capable of reasoning on a level beyond the capacity of an average engineer, when given the right tools and proper sandbox to test the ideas in (akin to our imagination). Also, LLMs perform the best when reasoning about things that were not in the training set. E.g., they write much better code in languages they've never seen - because they're forced to do it slowly, verifying every step, instead of churning out answers instinctively.
1
@vincentvogelaar6015 apparently you do not understand how to use LLMs. They're not any different from our own minds - we cannot reason either, unless we use tools, such as formal logic. So give LLMs their tools - give them an ability to write down reasoning step by step, to verify the reasoning using formal methods (as in - make them write down the steps as HoL proofs or Prolog predicates). Give them a sandbox to debug the proofs, just like you do with any other code. Provide a critical loop to make sure they did not miss anything from the formulation of the problem when translating it to a proof. I'm using LLMs for solving engineering problems, and reasoning is a crucial part of it. Even very small models (like Phi-3) are perfectly capable of reasoning on a level beyond the capacity of an average engineer, when given the right tools and proper sandbox to test the ideas in (akin to our imagination). Also, LLMs perform the best when reasoning about things that were not in the training set. E.g., they write much better code in languages they've never seen - because they're forced to do it slowly, verifying every step, instead of churning out answers instinctively.
1
@vincentvogelaar6015 youtube keeps deleting my answer for some reason. I'll try to make it shorter: in order to reason, LLMs must use tools, just like we do. Writing things down, visualising them, applying formal logic, you know, formally - and it all works really well with even the very small LLMs (like Phi-3). And LLMs perform the best on knowledge that was not in the training set, it's forcing them to slow down and reason in very small verifiable steps instead of generating an "intuitive" answer.
1
@vincentvogelaar6015 how is it not a reasoning, if this is exactly what humans do? With the result at least not worse, but often much better than human reasoning. You just give this autonomous agent loop with its toolbox a single, vaguely worded goal, and after some time and a lot of very small steps it will create a solution. Fully autonomously. Right now I'm using LLMs to design mixed signal circuits and mechanical parts this way. Just give them the goal, and they do the work. Including finding the datasheets for the required parts, creating and running simulations for things they're not certain about, etc.
1
@vincentvogelaar6015 see the Voyager paper for a rudimentary example of such a self-improvement. No need for further model training - there are more efficient ways of storing the accumulated knowledge.
1
There is never "too much calculus". The more - the better. But of course code monkeys will never understand why it's so important. You sound truly pathetic and revolting. I hope I'll never have to work with someone like you.
1
@ArchetypeLuna "clean code" is a fad popular among the lowest of the low code monkeys, not an essential skill. Unit testing is an obvious thing and details depend on the specific toolchain, so should not be taught in a generic course. It is better to teach formal semantics and proof theory instead, for there's too many code monkeys who put too much faith in unit testing where a formal proof would have been a much better guarantee.
1
@c00ckiez for the "real skills" you should have went to a trade school and then do an apprenticeship. Higher education is for the adults who can acquire the mundane lowly "real skills" on their own once they're given a fundamental knowledge and a robust framework for learning. You're pathetic, and I'm really glad I'll never have to work with someone like you. Luckily, all my co-workers are PhDs at least. You have zero chance of ever understanding why Laplace and Fourier and the rest of calculus would have moved you to a totally different level from the code monkey level you're permanently stuck in now. And you're a perfect example of why we have a shortage - too many useless code monkeys, and not nearly enough people who had guts to actually learn calculus.
1
Developers who do not use formal verification for critical systems are trash developers. There are no excuses. Ignorance is not an excuse. "Hard to find competent developers" is not an excuse. This industry is teeming with the wrong people who should have never been allowed anywhere close to programming - all those self-taughts, bootcamp "graduates" and such.
1
Complex-schmomplex, they built Horizon in a Visual freaking Basic 6, without ever even thinking about boring details like ACID. It was doomed to be a failure through and through, not just a "complex system with a few bugs".
1
@tongobong1 nah, a lot of companies are willing to pay well above the market just to get a mildly competent developer. Yet they struggle, there's just not enough developers and a lot of them are not primarily motivated by money so cannot be easily head-hunted out of their current jobs.
1
@kzelmer what a dumb take. The shortage is real and measurable.
1
@benisrood on a fact that it is very hard to find a single competent engineer, that companies viciously poach from each other, that recruiters call all the time and sound very desperate and offer obscene money. Everyone I know struggle hiring engineers.
1
@benisrood you can live in Barcelona and work for a company in the UK, or Switzerland, or USA, or wherever else pay is much better. This is exactly what I do, living in Barcelona on a UK-level income. Also, startups that I know in Barcelona offer way above 60k.
1
It is a systematic problem. If the industry is regulated, then both management and developers will have their hand tied, they'll have to do the right thing even if they really really want to ship crap to market asap. Incompetent developers are a problem - and regulations will weed them out. Incompetent management is a problem, and regulations will limit the damage they can cause.
1
@WhyteHorse2023 this is why this situation is so absurd - overproduction of juniors led to an unsustainable drop in average quality, made it nearly impossible to find those who are worth even trying to train, all the while experienced people retiring with nobody to replace them. If the junior pool was not so diluted with the wrong people it would have been much easier to train them into experienced engineers.
1
Many languages, some are good, some are mediocre, but none as awful as JS. I use C, C++, various flavours of Lisp (including Scheme), Tcl, Verilog, ocaml, and a few more, including even Fortran. Nothing is as ill designed as JS.
1
@bernardd yet, Borland Pascal compiled instantly woth no visible delay on an 8086.
1
 @PavelHenkin you could always safely distribute a hash. They were just some lousy code monkeys, they built a system on top of f-ng WiFi, obviously they never considered anything at all.
1
but imagine how fun it'll be if you'd do all these things with an actually capable language that does not cripple your creativity? Only people who had little exposure to the other languages and programming models can genuinely believe JavaScript is a right tool for expressing their ideas.
1
So, you really want to try to prove that there is no shortage? Really? Find me a few engineers who are not a total crap.
1
@arcomarco7131 Shortage is real - there are no engineers. How else can I prove it? It is far easier to prove an existence than non-existence. It takes on average 5 months and a hundred of interviews to get a new engineer. While you can easily prove that there is a lot of them, supposedly, readily available. And it is an experience of everyone I know. Less than one in a hundred is worthy of talking to, even less are actually hireable. And yet you insist that there is no shortage. Are you surrounded by hungry eager engineers who will switch jobs on a whim if offered an above the market compensation? Where is this paradise of yours?
1
@awmy3109 oh, how cute, to assume that IO is always a bottleneck and that you do not pay very dearly in performance for crappy abstractions and abstrraction-enabling features like GC and duck typing. You do pay, a lot, in orders of magnitude of performance difference.
1
@awmy3109 also, no, it is not a "fact" that IO is somehow always a bottleneck. This is something only a narrow minded web coder can say, who have no idea how huge the world outside of his pathetic web stack is.
1
@awmy3109 somehow my long answer got lost. Again: GC, dynamic typing, virtual calls are abstraction-enabling tools that harm optimisation and cost a lot in runtime.performance. This is why your JS is orders of magnitude slower than my C++.
1
@awmy3109 right, what a great argument! Since only JS is available in a web browser, JS is somehow now a great and performant language. Simply for a virtue of monopoly. Nice. But this is not what OP was talking about. This is how people with experience from outside of web see the web ghetto. Any time we take a closer look we recoil in disgust. This whole web thing is a pile upon piles of utter crap, and it should have never been like this. Also, people like you are very often guilty of not even trying to think if they really need to build a "web app". Turns out, very often they should not, but they still do, because this is all they know. A lot of time a native application is a far better solution. Even more often, no UI at all is even better, yet, the people with such a severe professional deformation fail to see it. Now, you claimed JS is somehow performant (we know it's not, not even close). You claimed you don't pay for abstractions in such a language. Wrong again, you pay dearly. Admit you're wrong and stop moving the goal posts.
1
@rentefald You most certainly can write game engines from scratch. If you believe you cannot, you must stay as far away from programming as possible, as you're a part of the problem of massive overengineering of pretty much everything.
1
really? Have you ever seen any proper dependently typed language?
1
@salvatoreshiggerino6810 Ada (SPARK) is what both pays well and got a really good type system. And projects tend to be far more exciting than the typical corporate stuff - automotive, aerospace, other kinds of high reliability embedded...
1
No. There was never any Java in Netscape. They just wanted to jump on a hype bandwagon, as Java was becoming popular in the corporate world.
1
The pre-winter AI definitely could fix it. You know - formal proofs, SMT solvers, etc. All the stuff modern so-called "engineers" who are likely nothing but some bootcamp "graduates" know nothing about and have no mathematical background to even start comprehending.
1
Nah. There are formal methods and well established design protocols that would have highlighted all such issues on a design stage. But the developers were incompetent and should have been flipping burgers instead of being anywhere close to any real world engineering.
1
@charlesd4572 there is nothing wrong with flipping burgers - it's a decent profession, and it is far more suitable for the self-taughts and bootcamp "graduates" than engineering. There were no unknown failure points in this system. Like, in every distributed system every link can go down. Obviously. "Engineers" who fail to use formal methods to evaluate all possible failure modes (based on known individual component failures) and their consequences are not engineers and should not be allowed to design anything at all. I've been building fault-tolerant systems for decades. It really hurts to see what abominations people build when they have no engineering background. Cache is indeed the wrong solution here - and using a WiFi for anything even mildly mission-critical is an outright crime. A fail-safe design would have had multiple communication channels and a quick way of detecting the issues with them.
1
@charlesd4572 there were no unknown failure points in this case. All were pretty obvious. Yet code monkeys failed to do anything at all with them. Do not try to find excuses - there is no excuse for doing such an awful job.
1
@charlesd4572 stop trying to find excuses for bad engineering. There is no excuse. They used wifi ffs, they clearly set for failure from the very beginning.
1
@charlesd4572 of course you can. There is an entire engineering discipline doing just that. Of course there are always unknown failure modes, but you can always design a system tolerant to all the failure modes you know about in advance. Have you seen aerospace and medical designs? Even automotive?
1
@martinjungmair705 software had a lot less bugs. A lot better usability. Was a lot more responsive than the current overbloated ill-designed software. And had all the features end users needed.
1
@martinjungmair705 of course there was a lot of crap software back then. Pretty much everything on PC was awful (and now is even more awful), with an exception for Borland products. I'm talking about the proper software (Unix, VMS) of that era. In particular, everything from Microsoft was absolutely awful, maybe with an exception for the first Windows NT release.
1
@ContinuousDelivery to add on the topic of continuous delivery of the embedded solutions - I found it very useful to have a simulated analogue hardware-in-the-loop. For example, for the aforementioned BLDC controller, we're using a patched ngspice version that supports mixed signal time step simulations, so we can run firmware in a loop with a simulated analogue part of the system (with even a load being simulated). Every change to firmware therefore runs through a set of tests, including breakdown scenarios that cannot be tested on a real hardware (well, they can, but only once).
1
 @almari3954 if you don't have any interrupts, and just poll all sensors in order, your FSM is always deterministic and allows a precise timing analysis.
1
@almari3954 if there are no interrupts, FSM is deterministic. There is no problem.with such "concurrency".
1
@pompiuses there's a caveat - Java is best performing only if you write it in such a way that GC does not have much to do. I had to deal with a few Java real-time GC implementations, it always boils down to writing in a very restricted subset of Java and continuously questioning - why the heck we did not just use C++? As for thoughput performance - yes, JRE is one of the best, and GCs available for it are very mature.
1
All the antivirus companies are just a security theatre and a scam. Microsoft should have never caved to them, AVs must be pushed out of the market.
1