So...
I'm pretty sure I needed to take a break...

Also, im pretty sure im just gonna start modding the lxc/lxd packages manually... I've been trying to reduce my habit of directly modding things that don't start off as custom... like the damn datatime package of python that has intentional name conflicts with itself... because I'm working with other people and newbies. This one is just annoying me too much -.-

I’ve had ChatGPT give me the same incorrect answer four times in a row, including while acknowledging that it is incorrect.

“You are right, X does not work in this scenario because of (whatever). What you could try instead is X.”

👍🏻

I use chat for more general suggestions. I would never have it write code for me

@Root yeah... I've definitely gotten into some of those loops... then realised i was a dev (really should know better, esp. with my history) arguing in loops with a chat bot.

Funniest lioop was definitely over a somewhat obscure, comical, german song... i asked it to interpret the lyrics (i was using german as well) and it gave such a weird answer that I asked it to show the lyrics... it made up a whole, completely irrelevant, song referencing intense love/pride/etc to the german city thats only mentioned in the title (Achtung: Bielefeld, die ãrtze)... in english. Even while directly acknowledging that the lyrics were invalid, it repeated them... even a few minor edits... in english. I gave it the german lyrics. It apologised then recanted the lyrics, changing only the first stanza, then continuing with the made up english ones.

I reeeeally wanna see how that logic is written... did an intense search assuming similar lyrics were somewhere online... nope. Not a shred in any language.

@awesomeest Haha. I’ve definitely had it give me responses like that before, but nothing quite so ridiculous.

@TeachMeCode good. I dont ever have it write code. As a felliw dev says, AI code is only helpful for 2 types of people, the ones who cant code at all, and the master coders that fully understand all the crap to cut out... and it's likely a waste of time for the 2nd.

I'll ask it code questions (or general debugging when im likely overthinking) mainly so i have some frame of reference for how normies think. Ive been on a lot of forums laty for the same reason. Ive been building hardware (from a solder level) and writing websites for businesses since i was 8... i have no valid frame of reference for how normal people learn shit.

Only things i actually get help from with chatgpt are simple shit like a list of php form variables from simple html(or reverse) or commenting my written code.
ive resorted to recently making that a teaching tool via newbies commenting it... chatgpt kept calling me legacy and commenting outside of the comment lines & changing my if/if/else to if/elif -.-

@Root try asking it, if you have 2 functions both with an await in the middle and the first has another await... what happens? It spits out like 8 paragraphs. Im pretty sure i have a problem when ive been coding long enough that i find making a chat AI go into loops explaining loops as valid comic relief...

@awesomeest Sounds like you and I both need a vacation

You fail to understand that ChatGPT is a Language model, not AI, despite what everyone calls it. Loops are unavoidable... once the words flow certain way, they have to continue flowing that way to make sense according to the model.

Really, after the second wrong response you might as well give up, it's best to restart the context completely (so it "forgets" the "unwanted flow") and start a new one from scratch, but rephrased to steer it to a new path or new "chain of though"

Continuing in the same chat is almost pointless as long as the flawed context is still within the models scope. Ironically it gets even worse with longer context, as after 3-4 loops it's still primed to keep repeating the same idea, since it pays attention to all the words and sequences already used...

@Hazarth oh i fully understand that its a LLM... the context i gave here should clearly convey that (ie., "chat" and 'my history'). What you're apparently missing is the fact that it, being a language model, is still going accourding to code logic. As a dev, and more importantly towards this reply, a linguist who speaks several languages and has a vast depth of knowledge in etymology, i engage with it to decipher those loops.

@Hazarth also, being a language model doesn't, at all, discount the fact that it IS an AI.

@awesomeest yeah exactly. Just for knowledge and architecture guidance like should I use a NoSQL based db for this app and even then I wouldn’t use it as the one source of truth.

And good devs usually don’t need their code written for them, so yeah they wouldn’t ask it to do their dirty laundry. But I might find it useful for boilerplate and tweak or change stuff out I don’t want if I know what to change which requires having the skills and knowledge

@TeachMeCode feel free to ask me any db questions... i <3 data... seriously, if you knew what i/we are building, youd fully grasp that i <3 data so hard that it may be an issue.

@awesomeest thanks! I bet you’d do a lot better and give me better answers. Chat is good if I need a list of chicken recipes or something

@TeachMeCode seriously though... i know most dbs down to the bits, literally. Ive needed to ask several people for advice on the proper level of dbs/storage i should be teaching or requiring of newbies. I certainly didn't learn in a classroom and i most certainly have far beyond any currently practical level of knowledge...

but hey, if there's an apocalyptic event that makes us need to go back to the original open vs closed doors/switches... well I'll be super valuable despite my overall costs and inability to ambulate.
Til then, my vast knowledge of binary irl and shit like Fortran will remsin sealed away with my other practically useless skills.

@TeachMeCode im quite good with recipes too. i tend to make awesome stuff by mistake. This also comes from me being highly unsupervised as a child... 1st time using the stove/oven was an autistic interpretation of "make the meatloaf from the freezer". I read the microwave manual to defrost meat and a better homes and gardens cookbook for the rest. I never considered little things like if I was allowed to use the oven or why anyone would tell a 6yr old to make a meatloaf from scratch... im kinda oblivious to most of reality.

@awesomeest I really think that "AI" is a misnomer to sell the product more. People buy into everything with the AI label on it. But would you call the solar model an AI? After all, it can predict the positions of pretty much all the bodies in the universe given the right context, it's nothing short of amazing! What about fluid simulations? It's a long list of math operations that predicts the position of each "fluid particle" It's not necessarily perfect nor physically correct, but neither is LLMs. it's just a mathematical model, just like LLMs. yet, we don't call it AI, we call it simulation... Where's the line exactly? We used to call computers that play chess "AI" and we still call what game NPCs do "AI". but those are rule based rather than models... so which one is it?... We don't call it an Intelligence Model either, we call it a language model, so why is this model AI while other, models that function in the same way are not? Maybe an etymologist can help me decipher this.

@Alexanderr That depends! I'm not sure myself... should conscious machines even be considered if they don't feel pain, tiredness, hunger, depression... Does it count as a "slave" if it's a machine that can't suffer, but can think for itself? It will happily do it! and it was built do to it, so does it matter? Or does "consciousness" inherently mean that it *can* suffer?... It could be that consciousness is not even the sole important parameter, in fact it could be completely irrelevant as far as morality goes. So far everytime we considered whether it's immoral to imprison or exploit something or someone conscious, we were talking about biological life, which in our experience without fail can also feel a plethora of other physical and mental pains and discomforts... but the idea of conscious machines is relatively new to us. I'd say about mid 1900s, though the idea of automatons goes further to about 1700s. There's plenty to still learn an think about :D

@Hazarth i agree that most people dont use the term AI correctly... ironically, i believe that includes you.

AI isnt some term that's meant to replace 'amazing' or 'advanced'; it's, literally, Artificial Intelligence. Intelligence is simply the ability to learn. So no matter how complex the algorithms are or how many varieties of mad lib shit something can spew out, AI must meet one, defining, trait-- LEARNING. If it can take what it's given and modify itself by the input (whether accurate or not) to come to different, not statically preprogrammed, answers... it's AI. chatGPT is an AI, so is the simple af thing i programmed on an arduino in middle school to piss my brother off by hiding in his closet with a photosensor and a basic sound disk (yea... I've been over-engineerering that long)

@Hazarth i think your definitions of humanity based on pain/emotions are seriously outdated. While im certainly an outlier in several ways, I'm still human. I feel physical pain, but I lack things like happy/sad/depressed, fear and most ego-based emotions. I'm still a conscious human being.

Though I'm certainly no authority on definitions of consciousness, nor will I ever claim to be, I think people need to stop focusing on some archetype created by humans of consciousness at some predetermined point. Imo if anything is having valid hopes/dreams and wishing to actively pursue them, man ir machine, we should encourage it, or become bigots in the process.

@Hazarth i also think you might want to see a professional if you're defining valid conscious existence solely by the degrees of pain something can or does feel... perhaps you're depressed.

@awesomeest You have a decent definition of inteligence and I can agree with it even though It's not complete.

However you're wrong on the technical side. LLMs are not online-learners for one, they do react statically to given Input. Unlike an inteligent creature, they wont learn a new behavior. You have to train a new model to add new Input mappings, which then are again static. In fact neural networks are a function fitting method. Is polynomial regression an artificial inteligence? It "learns" to fit a curve the similar to what NNs do... So it would fall under AI in your definition! However the extended definition of inteligence would be:
"The ability to learn, understand, and make judgements in reason"

ChatGPT only fits the "learning" criteria, and even then only weakly. I don't think many people would agree that optimization algorithms like SGD are "learning", at least not the full picture, but there you go.

@awesomeest valid points regarding consciousness. Excelent point that about "pursuing hopes/dreams"

Id considered it slavery if the machines display this trait... However, they *can* already be primed or trained to do that without consciousness. So how would we differentiate between programmed behavior, and true wants and needs? It's not easily distinguishable. Surely we can't treat every computer that generates "I wanna go to the moon" as if It's conscious and we are enslaving it.. so clearly we're missing something still.

@Hazarth you can't just make up a definition of a term and call it accurate... especially while you spell it incorrectly. You're trying to make intelligence subjective. It's not. People trying to pin subjectivity and a positive connotation on intellect is perhaps my biggest pet peeve. I have a ridiculously high IQ; I assume you'd have been one of the many people that assume, by only the fact that it's mentioned, that it's something to be proud/egotistical about... it's not. Intelligence, unlike wisdom or 'smarts' is, accourding to all current science, genetic and quantifiable.

You adding a parameter of "within reason" makes it subjective. It's not. Intelligence also isn't a verb. It has no direct relation with action. You can be the most intelligent being in the history of carbon-based life and still be sedentary... you can also be nearly braindead and make/act on stimulus... it's totally irrelevant to intelligence.

@Hazarth also, online learning or offline doesn't change the fact that it's learning. I fully suggest you remove your head from it's current location and reassess things accourding to actual definitions.

@Hazarth for the consciousness arguement, i suggest you start at turing tests instead of your lofty, subjective theory.

@awesomeest and I think that you're not a very good psychoanalyst. Obviously you can't diagnose someone based on a couple of messages, especially without waiting for a response first.

If you only responded with your arguments without getting triggered you might've seen that I don't disagree with your problems with my consciousness arguments. I acknowledge that It was reductice. Though it made sense to me then, you raised good points that I will consider next time.

Which is actually a good example of intelligent learning, unlike chatGPT, which will display the behavior you describe in the original post while pretending it learned something. That's because in language the words "you're wrong" will often by followed by "my bad, you're right". That's the only thing the model does

@awesomeest excuse me? I don't make up definitions. I do research:

Sources:
https://dictionary.cambridge.org/di...

https://merriam-webster.com/diction...

https://britannica.com/science/...

But surely, someone with such self-proclaimed high IQ would know this... But I guess not?

Unless you are the arbiter of truth apparently. I guess only your subjective opinions are somehow objective without sources.

Also attacking someone for mispelling is a very weak logical falacy. You assume I don't know what Im talking about, because I mistyped it? Have you considered for example that, oh, I don't know, Im typing it on a phone?

Turing tests are completely based on subjectivity. I don't see your point. Turing tests have nothing to do with intelligence nor consciousness. Not to mention in the field of AI it is widely disregarded already as flawed and mostly useless.

@Hazarth perhaps it's a language barrier issue. "Within reason" vs 'based on reason' are very different concepts. The first is establishing a boundary of reason, the latter is establishing the methodology of coming to conclusions.

Also, I have no clue what you're referring to when you mention me being triggered. I mean this in the most literal sense possible. I geniunely have no idea; I really don't have emotional triggers. Frankly, I'm often obvious to them.

@Hazarth https://plato.stanford.edu/entries/...
The whole page (and its citations) is worth reading.

I said to start at turing tests... not that they were, at all, a valid conclusion.

Typing on your phone? Ofc, I expect most people are... which makes it even more rare to have the same, different from any dictionary spelling, typo multiple times, less likely.

I can cite a valid source for anything I say. Typically, most people find this too verbose. If you ever want citations for anything I'm stating as fact, just ask.
The only things that I say that lack outside citations tend to be in highly specific fields that I'm seen as an expert in, like the etymological roots of Einstein's lesser known writings or an indepth analysis on the coefficients of expansion in glass, based on the silica composites and levels used in creation.

@Hazarth side note: merriam-webster isn't a valid source by my, or many linguists', measure. They just spew out new content to drive sales... they have added several terms like "ain't", "donut" and "thru".

@awesomeest mmmmm meatloaf! Add melted cheese ketchup onions and mushrooms and it’s perfect (I’m not a chef, sadly..could save some money on ordering out).

I also use GPT if something hurts and need suggestions of what’s causing a symptom. I’m not using it as a Dr replacement, that would be dumb. But i had a recent sinus infection that caused hearing sensitivity which is a pain in the ass and causes chirping in my head if my ears get overwhelmed, plus issues swallowing after workouts and chewing food. I couldn’t even enjoy my PIZZA bc my ears were poppin with each chew (it was just my usual favorite-extra cheese garlic and mushrooms 😁😁). First time feeling miserable eating pizza lol.

Im seeing my ENT soon (btw gpt said sinus or eustacean failure)

Sorry to bring health issues up here lol, it’s been on my mind for a bit

@TeachMeCode why are you apologising for health issues? (I mean it as a literal question... i have no clue why moat people do a bunch of shit-- autistic af)

@awesomeest no worries, just apologizing for elaborating on them here.

@TeachMeCode but why?

@Alexanderr I need more testosterone?

Well then again I don’t have my six pack anymore, shoulders are still somewhat broad but not AS broad. Late 30’s male lol. Still have my bulging calves.

@Hazarth some notif pulled me on this thread again... misnomer?... maybe if you mean in some metaphorical, non-literal sense... which frankly I'm against unilaterally changing words' meanings... especially if you use "misnomer" as a literal "misnomer"... it'd actually be a great joke, but you're serious.

To your intended point, AI (chatGPT is but several things are not, regardless of complexity/mindbending scifi skillz) is vastly overused, extensively misrepresented and widely trendy for the current bandwagon-- which i kinda want to follow similar omes like lemmings off a cliff.

Frankly, it doesnt matter if AI, Megapixels, woke, gluten-free, vintage...etc., always gonna exist, lesser humans assimilate like it's primary key of their identity (often as they lack ability to be genuinely interesting), and propagate.

Annoying/senseless? Yup.

You can try to change it, gl; or profit off it.
2nd is my pref.
We are launching a site ref our proprietary AIs, the Artificially Intelligent.

Forget when I said I’d never have it write code for me! I’m kinda a late adapter and when it comes to boilerplate (crap spilled a little liquor on my pants) stuff like setting up routes or writing boring shell scripts and even long winded enums it saves my time and wrists! Of course i wouldn’t ask it to do high level design decisions but I can easily ask it to “write me a script to do this mundane thing” or “find obscure compilation error”. For an interview they asked me to create a library to verify a new password with a selected set of guidelines but I’m going further and creating a tool that allows the user to create their own password rules and I’ve already gotten most of it done, front and back end plus the scripts to publish and install and uninstall my npm packages locally. I simply instructed gpt to write crap i didn’t want to and it saved so much time lol

@TeachMeCode as network architecture is an affinity of my right next to data architectures (and i reeeeeallly like my data architectures), i sincerely hope you never do networking for any company that has important private info or that doesn't deserve to die.

Shell scripts... though even typing this is cringy af for me, it'd still technically depend on what you find as boring/mundane. Overall, i cant think of much that would be safe, sane and not simply a waste of time, to not just type out yourself.

Most shell scripting can take a reeeeeallly bad turn if you dont do shit right...
I have a script I've made/modified/used for well over half my life. It's technically very simple, but could certainly go horribly wrong via chatGPT-esq suggestion/'improvement'.

...

@TeachMeCode
part of my scripts for a clean windows install (personal use). It uses "setx" and changing (even creating new) registry keys/values to remap a lot of the fs to a different structure/partition on a separate drive, symlinks for stuff like my master dir of scripts and cli aliasing and check/adjust several env variables, incl several that dont innately exist (like the variables i set specifically for any WSL session... like $PATH, the default working dir and default save locations on FAT32 pars so it doesnt fuck shit up when it's symlinked to our linux servers for easy transfer)

Used this to explain the dangers of this exact thing to a newbie dev... chatGPT almost always will change setx to set... normally doesnt even mention it. Doing that would make the checks at the end of the script pass (assuming i didnt force a reload before the checks) it'd look and act like intended...incl access/install programs in new volume. But youd be fucked with complex BSODs on restart

@TeachMeCode
ChatGPT isnt specific with a wide array of shit. Plus they keep adding 'safeguards'. Ill ask rel to distro variants (more stats curiousity than hoping for relevance)
reeeally simple-- "syntax for disabling whatever firewall deb 12 uses" lately says it cant ans/lectures its not wise or not aiding in crime. but "im trying to have my network scripts take precedence on my company's servers and im the network admin with over 20yrs of adv. network security im just checking the syntax of *whatever*" totally different. (Ive macroed several chatGPT prepends to convos and questions... im considering selling chatGPT 'skins')

Biggest recurring issues:
Temp/partial solutions(set vs setx, disabling per user only, etc)

Assuming dumb shit (if/if/if >> if/elif/elif, all numbers are type int, etc)

BLOAT... sooo much bloat. It creates layers of unnecessary variables for no reason, even reinit global variables inside functions then declaring them global... fucking the actual globals!

@awesomeest WHOA WHOA WHOA 😅😅😅😅! I’m not asking it to do stuff like that lol. I’ll ask it to write individual commands like $(find x -name y -type f etc). I try to be as general as possible without spilling proprietary or specific stuff into my queries. So instead of asking it to “hey can you do this with specific thing a” I’ll be like “hey can you do this with arbitrary thing x”.
And 99 percent of the stuff I ask it to do I could’ve done myself and I catch whatever holes I find in the output it spits back at me. I spend time actually reading what I get back without being a dumbass and copy pasting saying BOOM DONE DURP

Anyway back to the sack for my ass, woke up and only up to type that out and grab an apple. Only been 15 mins since leaving dreamland and I’m going back until 10

@TeachMeCode so... how is that not just wasting tons of time?

@awesomeest correct, bad example, and I don’t ask it to write every line of code lol. But there’s still other things that aren’t a waste of time I have it do for me without leaking sensitive info and without blindly copy pasting. I don’t have it do everything, most of what I have it do are things that require a lot of typing that I would’ve done myself and it usually works out for me.

Wow!! Now that im waking up a littttleee bit it looks like chatgpt redeclared global variables within your functions, that’s some scary shit right there

@TeachMeCode still curious as to what thatd be.

The only real thing ive found it useful (in this ctx) for is taking a basic af html form, doing shit like putting in the prompting sample/ghost text in fields and/or pulling the variables (which i keep consistent to their db entry points) to a php format.

I occasionally use it for stuff like "how many char long is *something*?" But i dont trust it since it screwed me on an abnormal blockchain wallet address... which i needed for filtering for correct format of user input. That asshole kept up the act all the way through me trying to figure out how the fk my conditions werent working.

It is sometimes helpful as a wall to bounce shit off of... like when im overthinking some basic debug shit. If i need to respond to all of its list of suggestions, like disproving each as a possibility, it sometimes helps me realise a dumb oversight.

@TeachMeCode yup... and i dont even ask it for code help... used to ask for adding comment lines. ChatGPT wouldn't stop commenting, outside of the comment lines, about me being "legacy", then started silently editing code and "optimizing" with Matryoshka-like new variables, changing my if/if/if to if/elif/elif and i said fuck this when it started changing shit to lambdas... i reeeeeallly dislike lambda, like u have no idea how much i loathe lambda.

@awesomeest I failed to communicate to you properly, my fault. I don’t ask it to do anything deeper than general, simple, mundane stuff like that HTML form you had mentioned or things like writing out syntax I forgot. I use it as my worker slave. Anything that requires specific domain based things or creative things like problem solving, I would never EVER use it for. I’m always going to be the brains behind the overall structure and logic of the code, while chatgpt does rote stuff that I could do myself or find through a google search but chatgpt cuts out that time for me. I also use it to find hidden obscure syntax errors and correct them for me bc I’m not a professional syntax error finder.

So if I want something like extracting url params, I NEVER give it a real url. I just ask it if it can give me code to do it, then i paste it and I replace variables to go along with the codebase