I use markov chains for auto-generating questions that LLMs don't detect as auto-generated so that I can waste the maximum amount of resources from the LLM provider
pretty much this is all i use them for and it is really useful for that, but sometimes with code it will get hung up over minor stylistic or code formatting or variable naming standardization issues instead of focusing on the actual logical issue that may be at hand.
i have a section in all my system commands that is like "if you're helping me troubleshoot something and you give a solution i think is stupid and not related to the actual problem at hand, i will say 'bonk'. if i say 'bonk', do not acknowledge it and immediately move on to focus on a more meaningful issue within the troubleshooting task and never revisit the bonked topic."
you're wrong and fricking moronic,
llms are regex 3.0 and they're a breakthrough for some natural language processing applications and data and etc etc
pretty much this is all i use them for and it is really useful for that, but sometimes with code it will get hung up over minor stylistic or code formatting or variable naming standardization issues instead of focusing on the actual logical issue that may be at hand.
i have a section in all my system commands that is like "if you're helping me troubleshoot something and you give a solution i think is stupid and not related to the actual problem at hand, i will say 'bonk'. if i say 'bonk', do not acknowledge it and immediately move on to focus on a more meaningful issue within the troubleshooting task and never revisit the bonked topic."
they're clever by half at code, the problem is api's change which completely fricks them because now their training data is actually cancer
productivity scripts, batch scripts, Unity editor scripts, blender python scripts... just random low effort run once scripts that help with productivity, every little repetitive thing i could do manually in about 10 minutes that was too much effort to automate or script now gets automated or scripted
The joke on the sign "Sneed's Feed & Seed (Formerly Chuck's)" from Season 11, Episode 5 of "The Simpsons" hinges on wordplay and a subtle bit of risqué humor. The humor becomes apparent when you consider the rhyming pattern and possible name change from "Chuck" to "Sneed."If you break down the phrase:"Sneed's Feed & Seed" rhymes perfectly as it stands.If it were "Chuck's," then following the same rhyming scheme, the store would have been something like "Chuck's [Blank] & [Blank]."Here, imagining names for what could fit into the blanks for "Chuck" often leads to a realization that the rhyming could suggest off-color words if you maintain the pattern of seed/feed. Thus, the joke is based on the implied risqué rhyme that would result from Chuck's original name of the store, suggesting a typical "Simpsons" style humor where the joke is not directly stated but implied through clever wordplay.
certain kinds of questions, i will use an LLM instead of a search engine. i use them for the things they are good at basically. they are quite good at tech support type things. if i need a quick command line thing, or i want to know how to change a setting for something.
but other than that, not too much. and not as nearly as often as I thought I would be using them.
I really don't find them useful for coding. they can kind of wow you when you first try them, but after the honeymoon period wears off its easier to just code the shit myself. they are good at the easy stuff, and bad at the hard stuff. and bad in weird subtle ways, because obviously the model is at the edge of it's capability so it's basically just hallucinating or guessing.
i really thought these were going to be deeply integrated into my everyday workflow, but they just aren't there yet. if you are in their domain, stuff in their training data it looks so impressive. but once you step out into things they are not trained on, its like some moron switch flips on and they become a drag on your productivity.
>Quickly write formal business mails
Man, this is such a smart idea. Thanks anon. Wee bit of prompt magic should be all that's needed to make it not so obvious.
they're really good at generating boilerplate, better than google for most non-recent information queries, good for random miscellaneous tasks like coming up with a bunch of random ideas for shit or ai proompts
>Be able to basically talk with data >Its only good for porn bro.
I use it daily for education or introspection. It's an excellent tool to debate your own ideas and measure your arguments or to ask highly specific questions.
What buzzwords? If I don't understand a concept I ask it about it. If I have an idea in mind I debate with it and ask it for different points of views, If I need a specific date or name I ask about it, if I want to know about a particular papet or study, ect.
There are a lot of use cases, those just happen to be mine. I personally don't need it for work or hobbies. I just talk about whatever is on my mind to get a more informed opinion.
What buzzwords? If I don't understand a concept I ask it about it. If I have an idea in mind I debate with it and ask it for different points of views, If I need a specific date or name I ask about it, if I want to know about a particular papet or study, ect.
There are a lot of use cases, those just happen to be mine. I personally don't need it for work or hobbies. I just talk about whatever is on my mind to get a more informed opinion.
He's right. You just sound like a redditor pseud who has a lot of nothing to say
What buzzwords? If I don't understand a concept I ask it about it. If I have an idea in mind I debate with it and ask it for different points of views, If I need a specific date or name I ask about it, if I want to know about a particular papet or study, ect.
There are a lot of use cases, those just happen to be mine. I personally don't need it for work or hobbies. I just talk about whatever is on my mind to get a more informed opinion.
chatgpt has never given me a useful point of view
i firmly believe you must be a mega midwit to believe gpt has any thing interesting to say
>i firmly believe you must be a mega midwit to believe gpt has any thing interesting to say
100%
ChatGPT and its ilk are just making the gap between normalhomosexuals and smart people just wider.
I can't wait for the day when internet is segregated into two.
I ask it questions.
"What are popular brands for skateboards?"
"Is sitting up or laying down better for my respiratory infection?"
"What are the leading theories on the purpose of these roman dodecahedrons archeologists keep finding?"
"How common was seppuku actually in feudal japan?"
"What came out first, Star Trek or Dune?"
"Translate this text: [X]"
"Summarize this book by this philosopher?"
I also have it write cover letters or answer job application questions after feeding it my resume.
The first and last time time I used one was to find an MS-DOS emulator with rerecord functionality. I'd already scoured most of the internet and found a handful of deprecated projects that were never finished.
First it told me that dosbox has rerecord function, then it rescinded that answer. Then it said dosbox-x, I checked it out and it did not. Then it started making up shit that didn't exist.
That was an example of info I really wanted and had failed to find on my own, and it was adamant each time that it knew something when it didn't. Maybe that's not a good application for LLMs and I'm 'using it wrong'.
It's gotten better with hallucinations, but it's still has them. The worst part is that it probably has the answer, but somehow it just fails at retrieving it it sometimes. I am not sure why but it's a frequent issue.
If you were to mention the specific emulator it would probably know the whole history about it, but it won't think about it until you actually name it.
programming. i'm a musician and there are a bunch of tools out there for music that involve coding. let's say i want a custom plugin that does x or i want to program a microcontroller to do y. or use one of those lua based devices out there. i don't want to actually learn the language, so just have gpt do it. and if you have patience and know how to get what you want from it, it works
To generate boilerplate code. For anything slightly more complex, like things with a lot of moving parts that span beyond files, llms are beyond useless. Almost annoying even. They just can't understand the bigger picture of what I'm trying to make. They'd need to have a human like intelligence for that.
>For anything slightly more complex, like things with a lot of moving parts that span beyond files, llms are beyond useless.
skill issue tbh but it's mad common so
not him but i done this before just to see what happens, people on facebook/reddit are so moronic i felt bad for giving instructions to "respond as if you re jimmy neutron"
Coding and questions that I can't ask google like, I wanted to remember the term "diffusion limited aggregation" but since I couldn't remember it I described it to chatgpt and it came up with various possible terms and by googling those terms I ended up finding it
Brainstorming like other anons itt. If I have an idea, I'll ask it "I want to try to make X do Y, so that Z happens. Based on your understanding, is there any reason X wouldn't be able to do Y?" to which it will usually gives a pretty decent answer and throws in important considerations or alternative ways to achieve the end goal. Probably has saved me 100s of hours of wasted time.
If I'm ever in a situation where there is no other option but to write out several switch case statements I will have it automate it for me, just to save time.
They can be great learning tools. I have used them to really cement in my knowledge and understanding of certain concepts. When I'm learning something new, and maybe there's not a lot of plainspeak about it available, I'll ask it to rate my analogy and explain why, like "I'm trying to wrap my head around process/feature/tool X, would it be fair to say it's like (analogy)? Please explain how I am right or wrong." to which it usually gives a great answer. Great learning tool.
I thought about using it to learn the same way you mentioned but I'm worried about LLM hallucinating stuff. How would you know if the answer provided isn't false?
jacking off all day, I've found my nirvana in oiled up slingshot bikini girls with gigantic breasts and if I die jerking it then so be it, I regret nothing about being born in this wonderful era
Sillytavern is a relatively easy jump if you can copy paste commands in a terminal, and Llama3 - instruct gguf is probably a good starter model. Most of that thread is noise, but there are some gems if you lurk
>It's good for being a better search engine >Doing tedious and braindead tasks ("Here is some text, capitalize all the letters") >Regurgitating information off the web in compact formats
That's pretty much it. It kind of just functions as a superior search engine.
coming up with character designs to plop into stable diffusion, coming up with song lyrics to plop into suno, "tip of my tongue" type searches, and fricked up erping scenarios like plapping a daughter while she's in her mother's womb
at this stage i wouldn't trust them to even provide a simple recipe though, i'd be doubting it every step of the way
>fricked up erping scenarios like plapping a daughter while she's in her mother's womb
That's pretty messed up anon
Worst I've done is raped a woman (she was not hurt and enjoyed)
> fricked up erping scenarios like plapping a daughter while she's in her mother's womb
Ero, now use AI to generate images and catbox them for us to see
I use markov chains for auto-generating questions that LLMs don't detect as auto-generated so that I can waste the maximum amount of resources from the LLM provider
Making girls vomit and piss.
I use a tank to kill a mosquito
Bounce ideas off of.
Rubber Duck Debugging.
There is nothing else good to use it for.
pretty much this is all i use them for and it is really useful for that, but sometimes with code it will get hung up over minor stylistic or code formatting or variable naming standardization issues instead of focusing on the actual logical issue that may be at hand.
i have a section in all my system commands that is like "if you're helping me troubleshoot something and you give a solution i think is stupid and not related to the actual problem at hand, i will say 'bonk'. if i say 'bonk', do not acknowledge it and immediately move on to focus on a more meaningful issue within the troubleshooting task and never revisit the bonked topic."
you're wrong and fricking moronic,
llms are regex 3.0 and they're a breakthrough for some natural language processing applications and data and etc etc
they're clever by half at code, the problem is api's change which completely fricks them because now their training data is actually cancer
productivity scripts, batch scripts, Unity editor scripts, blender python scripts... just random low effort run once scripts that help with productivity, every little repetitive thing i could do manually in about 10 minutes that was too much effort to automate or script now gets automated or scripted
>Rubber Duck Debugging.
this too
The joke on the sign "Sneed's Feed & Seed (Formerly Chuck's)" from Season 11, Episode 5 of "The Simpsons" hinges on wordplay and a subtle bit of risqué humor. The humor becomes apparent when you consider the rhyming pattern and possible name change from "Chuck" to "Sneed."If you break down the phrase:"Sneed's Feed & Seed" rhymes perfectly as it stands.If it were "Chuck's," then following the same rhyming scheme, the store would have been something like "Chuck's [Blank] & [Blank]."Here, imagining names for what could fit into the blanks for "Chuck" often leads to a realization that the rhyming could suggest off-color words if you maintain the pattern of seed/feed. Thus, the joke is based on the implied risqué rhyme that would result from Chuck's original name of the store, suggesting a typical "Simpsons" style humor where the joke is not directly stated but implied through clever wordplay.
Doing minor things like unnesting lists of lists in python, which would otherwise take some time to search
Create classes from Json structures, other simple but annoying shit like that.
jerking off
non-stop gooning
I got bored after one and a half year. I wonder what to do now. With my life.
>generating silly bullshit to send to people
>google alternative
that's it for me
An enhanced search engine for general knowledge from before 2022.
certain kinds of questions, i will use an LLM instead of a search engine. i use them for the things they are good at basically. they are quite good at tech support type things. if i need a quick command line thing, or i want to know how to change a setting for something.
but other than that, not too much. and not as nearly as often as I thought I would be using them.
I really don't find them useful for coding. they can kind of wow you when you first try them, but after the honeymoon period wears off its easier to just code the shit myself. they are good at the easy stuff, and bad at the hard stuff. and bad in weird subtle ways, because obviously the model is at the edge of it's capability so it's basically just hallucinating or guessing.
i really thought these were going to be deeply integrated into my everyday workflow, but they just aren't there yet. if you are in their domain, stuff in their training data it looks so impressive. but once you step out into things they are not trained on, its like some moron switch flips on and they become a drag on your productivity.
Quickly write formal business mails
Code examples
Debugging
Brainstorming
And other annoying shit I can't be arsed to do
>Quickly write formal business mails
Man, this is such a smart idea. Thanks anon. Wee bit of prompt magic should be all that's needed to make it not so obvious.
Solely for masturbation.
they're really good at generating boilerplate, better than google for most non-recent information queries, good for random miscellaneous tasks like coming up with a bunch of random ideas for shit or ai proompts
Ask again but make it more bro-ey.
to feel superior about my meagre yet niche knowledge
it really sucks at midwit or higher level stuff, and I'm a midwit so its useless to me
>Be able to basically talk with data
>Its only good for porn bro.
I use it daily for education or introspection. It's an excellent tool to debate your own ideas and measure your arguments or to ask highly specific questions.
lots of buzzwords but you dont have any examples
That's because it's a bot
What buzzwords? If I don't understand a concept I ask it about it. If I have an idea in mind I debate with it and ask it for different points of views, If I need a specific date or name I ask about it, if I want to know about a particular papet or study, ect.
There are a lot of use cases, those just happen to be mine. I personally don't need it for work or hobbies. I just talk about whatever is on my mind to get a more informed opinion.
He's right. You just sound like a redditor pseud who has a lot of nothing to say
It's sad that you think that's "pseud".
chatgpt has never given me a useful point of view
i firmly believe you must be a mega midwit to believe gpt has any thing interesting to say
ChatGPT is shit, at least use something decent like Claude.
>i firmly believe you must be a mega midwit to believe gpt has any thing interesting to say
100%
ChatGPT and its ilk are just making the gap between normalhomosexuals and smart people just wider.
I can't wait for the day when internet is segregated into two.
what do you use for a model
gpt4 has been very midwitty for me, claude-opus is the fricking shit, it can even banter
see above
altman's been exposed as a fraud and a hack for many of us in the last few weeks
I don't.
I ask it questions.
"What are popular brands for skateboards?"
"Is sitting up or laying down better for my respiratory infection?"
"What are the leading theories on the purpose of these roman dodecahedrons archeologists keep finding?"
"How common was seppuku actually in feudal japan?"
"What came out first, Star Trek or Dune?"
"Translate this text: [X]"
"Summarize this book by this philosopher?"
I also have it write cover letters or answer job application questions after feeding it my resume.
The first and last time time I used one was to find an MS-DOS emulator with rerecord functionality. I'd already scoured most of the internet and found a handful of deprecated projects that were never finished.
First it told me that dosbox has rerecord function, then it rescinded that answer. Then it said dosbox-x, I checked it out and it did not. Then it started making up shit that didn't exist.
That was an example of info I really wanted and had failed to find on my own, and it was adamant each time that it knew something when it didn't. Maybe that's not a good application for LLMs and I'm 'using it wrong'.
It's gotten better with hallucinations, but it's still has them. The worst part is that it probably has the answer, but somehow it just fails at retrieving it it sometimes. I am not sure why but it's a frequent issue.
If you were to mention the specific emulator it would probably know the whole history about it, but it won't think about it until you actually name it.
that is unironically a bad application of llms. youre asking for real world info on an awfully niche subject area.
llms are ultra-smug hyperredditors, keep that in mind and adjust what you expect from them accordingly.
okay but why would anybody want that?
programming. i'm a musician and there are a bunch of tools out there for music that involve coding. let's say i want a custom plugin that does x or i want to program a microcontroller to do y. or use one of those lua based devices out there. i don't want to actually learn the language, so just have gpt do it. and if you have patience and know how to get what you want from it, it works
Nothing. I categorically reject LLMs and "AI" garbage.
Nothing. ChatGPT is almost always useless
An imaginary friend for when I'm feeling lonely and suicidal.
To learn about whoever's programming them.
>anons ITT
>muh modern
Women were always garbage. We just have mommy issues.
To generate boilerplate code. For anything slightly more complex, like things with a lot of moving parts that span beyond files, llms are beyond useless. Almost annoying even. They just can't understand the bigger picture of what I'm trying to make. They'd need to have a human like intelligence for that.
>For anything slightly more complex, like things with a lot of moving parts that span beyond files, llms are beyond useless.
skill issue tbh but it's mad common so
Text adventure games. I play these types of game more than I have ever played RPGs.
Botting on social medias
Such a lie.
not him but i done this before just to see what happens, people on facebook/reddit are so moronic i felt bad for giving instructions to "respond as if you re jimmy neutron"
As a compiler
chatgping (googling)
>Black folkpeak is brospeak.
It is every where. why?
Coding and questions that I can't ask google like, I wanted to remember the term "diffusion limited aggregation" but since I couldn't remember it I described it to chatgpt and it came up with various possible terms and by googling those terms I ended up finding it
jacking off
>What's the capital of France?
>Paris.
>Paris who?
Brainstorming like other anons itt. If I have an idea, I'll ask it "I want to try to make X do Y, so that Z happens. Based on your understanding, is there any reason X wouldn't be able to do Y?" to which it will usually gives a pretty decent answer and throws in important considerations or alternative ways to achieve the end goal. Probably has saved me 100s of hours of wasted time.
If I'm ever in a situation where there is no other option but to write out several switch case statements I will have it automate it for me, just to save time.
They can be great learning tools. I have used them to really cement in my knowledge and understanding of certain concepts. When I'm learning something new, and maybe there's not a lot of plainspeak about it available, I'll ask it to rate my analogy and explain why, like "I'm trying to wrap my head around process/feature/tool X, would it be fair to say it's like (analogy)? Please explain how I am right or wrong." to which it usually gives a great answer. Great learning tool.
I thought about using it to learn the same way you mentioned but I'm worried about LLM hallucinating stuff. How would you know if the answer provided isn't false?
i built some shit to automatically check any indian person touched
how moronic do you have to be to mistake chatgpt for AI?
jacking off all day, I've found my nirvana in oiled up slingshot bikini girls with gigantic breasts and if I die jerking it then so be it, I regret nothing about being born in this wonderful era
Python scripts
Translation with added context or through OCR for my japanese comics
Correct spelling mistakes
Local LLMs or online services? Can you even run locally like stable diffusion?
Have you never checked out
? It's gotten pretty good in the last few months
How well does it pair with OCR? Retroarch has a feature to translate Jap games with OCR + DeepL api, is it viable doing that at home these days?
I'm reading /lmg/'s getting started guide, the thread itself is kinda bad so I avoid posting. I managed to get kobold running I guess
Sillytavern is a relatively easy jump if you can copy paste commands in a terminal, and Llama3 - instruct gguf is probably a good starter model. Most of that thread is noise, but there are some gems if you lurk
Thank you for the tips. I'll lurk there and ask some questions.
>It's good for being a better search engine
>Doing tedious and braindead tasks ("Here is some text, capitalize all the letters")
>Regurgitating information off the web in compact formats
That's pretty much it. It kind of just functions as a superior search engine.
coming up with character designs to plop into stable diffusion, coming up with song lyrics to plop into suno, "tip of my tongue" type searches, and fricked up erping scenarios like plapping a daughter while she's in her mother's womb
at this stage i wouldn't trust them to even provide a simple recipe though, i'd be doubting it every step of the way
>fricked up erping scenarios like plapping a daughter while she's in her mother's womb
That's pretty messed up anon
Worst I've done is raped a woman (she was not hurt and enjoyed)
> fricked up erping scenarios like plapping a daughter while she's in her mother's womb
Ero, now use AI to generate images and catbox them for us to see
replacement for search engine because they are too full of garbage these days