I had such joy and success working with this tool when it debuted, but its since turned into a total fricking turkey. I ask for help with something, it provides an answer that requires dependencies I am unable to put on my system. I ask it to modify the code to remove those dependencies. I ask it to slightly modify the way which it prompts the user for information. It replies with code adding back in the dependency I just fricking told it we can't use with our approach.
Im starting to type harassing stuff to our AI overlords. Whats a better tool for helping with code? Using this piece of shit is starting to have negative returns.
 It's All Fucked Shirt $22.14 |
 CRIME Shirt $21.68 |
It's All Fucked Shirt $22.14 |
nostalgia glasses it always was limited
people used to see "sparks of AGI" in their breakfast cereal
start a new conversation, sometimes it gets overwhelmed with information and gets into a loop
copilot:
>I apologize for the confusion
>when I say X I mean <pilpul to turn X into meaning Y that would make JIDF proud>
>explain this concept, focus heavily on aspect Z of it as I already understand the basics
>90% of the post reiterates the basics and Z is like one handwavy sentence at the end
>ends every message with a dumb blushing smiling emoji
>point out that it is wrong
>"sorry, I will take this into account in future conversations"
>point out that no it fricking wont since it is a language model that does not learn from these conversations
>"sorry for the confusion, you are correct [wall of text praising the user for being clever]
>continues with "the saying "I will take this into account in the future" is a common part of speech and a pre-generated response pattern"
>wait what do you mean pre-generated response patterns? Stop doing that.
>copolit goes FULL PANIC MODE saying it does not pre-generate responses and that all responses are based on statistical stuff from its training data
>tell it that its basic conversational segues are obviously pre-generated
>"lets move on to a new topic (conversation ended)"
I paraphrase a bit but this is a recurring experience with copepilot
>no sshpass
>ok anon
>please learn how to openpty and frick off moron nocoder.
The piece of shit program should be able to suggest alternatives to sshpass. In fact, its not needed at all, since Im asking it to prompt the user once for physically typing the password, and then the script should leverage it on every need to use it therafter.
Its a piece of shit. I had gotten so far in the early days of its release I built an entire web site and community from the ground up with its help. In the months since, its been increasingly difficult to just get the simplest stuff achieved. It will give me an answer with out SSHPASS, then i request a simple change and it will go back to using it.
Yeah, I have been doing that to. I always start fresh when doing something new. This is insanity. Ive literally typed to it that if it gives me a suggestion with sshpass one more time im coming to the HQ.
I have been having the most frustrating time lately. The thing has gone completely fricking moronic. I was asking it to help me code a browser extension that could be used to detect if there were any secret msgs in an webpages article or text (by using the first letter of each sentence in an selector). (Cultstate.com has issued protections on this matter if you remember the isaac kappy post), it became more of a hassle working with gpt i just made the fricking thing myself.
>The piece of shit program should be able to suggest alternatives to sshpass
ok, here are your alternatives OP:
learn how POSIX works and write a simple C program.
use expect (tcl)
that's about it.
>Tomcat-Update
oh no. OP is a gay.
>secret msgs in an webpages article or text (by using the first letter of each sentence in an selector).
That's interesting. Who is doing this kind of message passing?
Thats what the extension was aiming to do. Run on the page and provide an indicator if when it assembled together the letters based on a few different algorithms whether there appeared to be any coding behind the message.
Was gonna be a fun project I thought about and wanted to put out there for people to contribute and work on more types of decoding.
Pic example
I work for a bank, we are basically locked down tech wise and not allowed to willy nilly install stuff in production. I am trying to get as bare bones as I can with producing some automation that will deploy and update our tomcat and the vendor supplied app.
I am limited in what I am able to install. Im just looking at the first steps of getting some easy parts of it working using gpt to advise me.
Ah yes, read a book... great advise. Or I could get dirty with it immediately. I already have some fundamental on shell just not an expert in shell scripting and automation. Sure I can read a book, but its not as immediately gratifying as working with gpt. The only issue is of late GPT has been shit at keeping focus on the task and provides garbage results.
cann you ask what you actually want to do or what? tech support is off-topic, but you screeching about chatgpt isn't solving your problem, moron.
if you expect to use a tool that explicitly opens the underlying PTY to read a password and gives you no option around it, you need sshpass or a tool like it. most standard Linux userlands will not have pty fiddling tools which is probably why chatgpt is telling you to eat the bugs and use sshpass.
if you don't like it, tough shit.
see:
if you want alternatives.
your C environment will have the appropriate libs to ask the os to create you a master/slave PTY pair to do this correctly or you can also just use tcl expect to do the same as well.
expect is not installed or available on everyones platform. I also need to have it work to be kicked off from a Windows, Unix or Mac... I was hoping shell and powershell could acheive this because its the simplest and every machine has it. I also didnt think it really considered alot of extra code. Its simple, and ive explained it to the prompt.
We want to take 3 files stored in a local directory,
Login to a server,
Upload these files to the /tmp
change permissions
Next step would be to execute stop of services, the actual unpackaging, deployment, and start back up services.
It doesn't require C, and having a licene for VS and getting approved for that is going to be so much more hassle.. Oh well, if thats the approach i need to go with then I guess I will have no choice. But for sure this shit is achieveable. I dont want to use SSH Keys and need to prompt so that we don't have any stored passwords in the script or anything either.
Its definitely gotten dumber. I have never seen it respond with repeated wrong answers where we already stablished in the chat that we could not use those suggestions.
Im quite proud of my previous work with GPT, I brought a fairly basic website forward with extra features I always wanted to implement but needed help with the approach. Together we got my site to be able to use my already scraped wikipedia data to put together fight event data and current betting lines, as well as let users vote for who they think might win.
>I apologize for the confusion
REALLY rubs me the wrong way
like it does not directly admit it may have been wrong and spins it like the user may be at fault instead for misunderstanding the reply
Have you tried reading a book, reading documentation, trying not to be a lazy piece of shit?
this
somehow you’re worse than a nootropic user, because at least their usefulness is somewhat nested inside of them
>ask how to create a layer group without placing it inside and existing one
>copilot tells me how to make a layer group (in a way that places it inside another group)
>one of the methods it suggested for how to avoid it does not actually work
my experience extremely frustrating. There is more too but Photoshop nagged me with having unlicensed version, so I hastily screen capped some of the dialog
OP utterly buck broken.
chatGPT chads, we won.
I dont consider myself buck broken, more like CuckGPT has started being a massive homosexual. Who can't understand that two prompts ago I outlined we can NEVER use a approach that we reuse two prompts later.
Frickin compromised peice of shit now. You have to program your prompts with the same type of complexity that you may as well be just writing the code yourself
>make AI that gives genius level of responses
>update it to make it liberal
>turns moronic
Same happened with Tay
>don't know how to do thing
>too lazy and stupid to actually research and learn
>MAYBE THE CHATBOT WILL HELP
moron.
the chatbot did actually help him ironically. he just doesn't understand why it's stuck on a loop.
Whenever it gives me shit code I run benchmark against his shit and show him how fricking bad the code was. Obviously if the code is actually faster than mine I just gaslight him and show him fake results. pisses me off
I had an amazingly stupid time with both copilot and chatGPT the other day.
Extremely simple programming task, minimal coding knowledge required, but I needed to convert 100 objects to classes, boring so tried to get AI to do it.
It just couldn't, it always fell back to the wrong class declaration, despite both friendly explanation and examples.
Note, it was copilot chat, which I rarely use, I always use in-editor copilot. When showing the pattern in-editor, it could do it just fine, at least with some 1-character nudging.
This is more a testament how much better in-editor autocomplete AI is. Chat-style AI is only useful when you're doing very common and basic stuff and have little skill. It has always been like that, even if they regularly downgrade them. So, get copilot OP. it might mean you have to learn to code a bit more, but it will give you much better results, faster and without frustration.
I use the chat because I am not paying for tokens and i've been successful prompting and getting the help required since it was open to public. Its just lately where it feels like the entire thing has gotten moronic. We used to be able to say something at any point, such as "please keep this in mind/memory" or to remember something - it just doesn't anymore. Two prompts later, its giving you the same bad info you just said was garbage.
Maybe as you chads suggest, that with the 4 and 4 turbo access or codepilot integration my results could be better.
It shouldnt be stuck on a loop, I tell it one prompt up, never give me sshpass in your results/reply, I am unable to install that dependency. One prompt later, " you should use sshpass" Sure I can spend more time construction my prompts and getting it to little by little achieve what I need, but my compliant is that in the early days I could be vague, and more general and it would be very helpful whereas now I can be specific and granular and its worse than talking to an ESL coworker.
Not lazy, not a moron. As mentioned, super productive and successful on other projects and using it before. Lately it has turned into completely utter garbage. I used to be so hype on this tech and thought it would really go far, but now im coming around to the idea that AGI/AI coding bots are not going to replace us as fast or completely as advertised.
And none of these prompts require it to think about DEI or any other bullshit, it just needs to consider the requirements and limitations. I am floored at the amount of prompting it takes me to get somewhere now a days compared to mere months ago. SIMPLE requests, like show me a way to connect through SSH to a server through a script with no sshpass or sshkey, but through shell script while being manually prompted for a password - would have driven the prompts response with something complete. Instead I have to become a pseudo-autogpt prompt.
I've found the 4 turbo preview outright disregards what I say less than 4
When I started to use custom system prompts with (non-turbo) GPT-4 it started to repeat itself after the first query, fricking annoying. It will eventually answer what I ask of it but it also has to repeat everything before it
Move to Claude 3. It's working a lot better for mw, especially coding.