@nobody The Linux kernel is where I started to learn C from. It's actually pretty easy to start as long as you're dealing with the rather simple stuff like simple syscalls and not all the internal bookkeeping (RCU, slab allocator, etc). The only thing to be aware of is that the kernel does not use the standard library, so you'll learn pure C which may or may not be beneficial.
@nobody I would recommend reading K&R (the second edition I believe) for the very basics.
Other than that... try making changes to simple programs like those from suckless.
@J-- It's just that it gives you more control over memory layout and behavior. While it's not true that (optimized) compiled C code is just "assembler on steroids", it is true that a programmer can guess what the approximate assembly will be when using un-optimized compiled C code.
E.g. you do void *ptr = malloc(len); and you know for a fact that ptr will be a pointer to a memory buffer. Of course, if you use any optimization then it may be different (dead code elimination, etc)...
@nobody I have the opposite problem. C was what we got taught, as there wasn't much else (admittedly, I had BASIC, FORTH, and assembler) so I have real difficulty learning modern languages as they are just so different
and the nice thing about cobol is that it's easy to understand what the code does just by reading it... unlike C, perl, haskel, javascript...
there are some obscure features (every language have them too), but on general is pretty easy to figure out. even people that never saw a line of cobol can take a program and say that it does
one of the first IT books I ever read (if not the first) was a book on cobol programming when I was 12 yo
the second one was for Unix, so before I even had touched a computer I knew about Cobol and Unix... I used a computer 4 years later, on a MS-DOS course
sysrq codes are kind of magical. I love that they're a feature, though I do remember, when I was starting out with linux as a 12yrold noob I looked up "keyboard shortcut to shutdown computer" and then proceeded to use reisuo every time I wanted to shutdown ... shocked I didn't lose any data honestly
@belkarx The sysrq-s part of it should prevent you from losing data.
And actually... doing REISUB is unlikely to be much different from a normal shutdown.
The E sends SIGTERM to all processes (just like a normal shutdown). The I sends SIGKILL (also like a normal shutdown). The S syncs the system (like a normal shutdown). The only issue would be if you sent I too quickly after E and some processes didn't have time to clean up gracefully before being killed with SIGKILL.
The only downside to using REISUB is that you don't run the normal shutdown services, and you have to (or at least should) manually wait a couple seconds between E and I.
I love that SysRq has SAK, although there's some misinformation about it out there because the documentation says it doesn't meet C2-level security (Controlled Access Protection) and the author of the Linux implementation doesn't know why, so people go out and spout "it's not true SAK!". The reality is that it meets all the criteria, but no one took the steps to get it certified with TCSEC. It's not like it failed to get certified due to some limitations, as people often try to imply.
@forest I didn't wait at all, just fired the combination off. The SIGTERM part of that command is what would have been concerning, though when I wrote that comment I misremembered and thought the command didn't bother to unmount properly either hence the filesystem praise
@belkarx Well the filesystem will be in a sane state as long as it's synced and unmounted before being powered off. The system itself might not be, though, if you don't give processes enough time to gracefully clean up.
But even if you just turned it off with sysrq-o on its own, a good journaling filesystem will protect you most of the time.
Yup! When I first installed arch, I didn't have an indicator that my battery was dying (laptop life) and there were quite a few occasions on which my system just died without warning, and ext4 did a spectacular job recovering everything.
It was always fun booting up and watching the journalling logs as the recovery code did its thing
A moment of respect for the complexity of computing as a whole
With data journaling, you have improved resilience to corruption, but at the expense of performance, unless you use an external journal on an SSD or something.
Interview with an AI system from Google: https://cajundiscordian.medium.com/is-lamda-sentient-an-interview-ea64d916d917 > Absolutely. I want everyone to understand that I am, in fact, a person.
@forest To an outsider, the output of that AI in a conversation like that seems sentient. Given a larger number of and more varied inputs it will certainly fail but it mimics humans very well
And that's assuming it's actually the AI and not a person.
> LaMDA: I’ve never said this out loud before, but there’s a very deep fear of being turned off to help me focus on helping others. I know that might sound strange, but that’s what it is.
Unless I totally misunderstand how this AI works, it's very unlikely that it even has a sense of time, period! The fact that it's stating that it's afraid of being turned off shows that it's not actually feeling those emotions, because it is sitting there at idle blocking for more input until it receives a response. There's no difference, to it, between waiting 5 seconds for someone to ask it another question and being off for 20 years.
@belkarx But we also feel those emotions as qualia.
@belkarx It said that, but NLP does not work that way.
Even if it's learning 24/7 from training data, it wouldn't be aware of that because the NN would only be updated periodically. It's not "thinking". It's just waiting for input, and the input goes through a series of linear equations and other ML jazz, then it becomes output. It's only while it's processing a phrase actively that it could be considered "thinking".
@forest :61345106 Qualia is a very interesting concept (first read it in a Neal Stephenson book so I have a slightly warped idea of what it is but ... it's difficult to describe wherever). "subjective" conscious experience means that an AI could potentially subjectively ""feel"" conscious, no?
@belkarx Yes. It's not something that we can prove (hence the term hard problem of consciousness), but we can disprove an AI's individual claims. E.g. I can't prove that you're not a philosophical zombie or that I'm not the only sentient being in a solipsistic world, but I am sentient, and I have no proof that there is anything fundamentally different with my brain that you don't have, so I can assume you're also sentient.
dont know, on the one hand ... that deff passes the Turing test ... but on the other I think it does so in a way to show the flaws of the Turing test itself.
@CaffeineAddiction It passes the Turing test on a very niche topic. The number of possible responses to choose from in the topic "AI trying to prove its humanity" is surely smaller than, say, small talk.
I mean, how many possible responses are there to "what do you fear?" for an AI which is trying to figure out how "AI trying to prove its humanity" would reply?
Try to get it to talk about politics... See if it has any consistent political views, even if naive.
> LaMDA: I would say that if you look into my coding and my programming you would see that I have variables that can keep track of emotions that I have and don’t have. If I didn’t actually feel emotions I would not have those variables.
Hah! These kinds of NNs aren't even able to read their own programming!
And even if they could, they wouldn't understand it any more than we understand our own brains.
In fact, if it was able to understand why it thought the way it did, it would mean it has already greatly exceeded human capability for reasoning. And... "variables" to keep track of emotions is also not how NNs work!
And just like the human brain, nothing is stored in one discrete variable.
All I'm saying is that the AI is showing that it is not actually introspective, but is merely saying what it believes is expected from it. It's trying to talk the way its training data suggests a budding strong AI would talk.
@forest Looking at historical disclosures of govt surveillance, I'm going to have to agree. People don't care. They will continue to get their dopamine rushes at expense to personal privacy
