I don't find it useless, I use GPT-4 for little things all the time
Doesn't seem like there's anything better on the horizon even from OpenAI though, I think Yann LeCun is right that language models are a dead end and this is about as good as they're gonna get
The concept of GPT models is only about 6 years old. GPT-2 came out in 2019. ChatGPT is only one year old. I think it is a little early to say it's joever for large language models.
I think we can do better if we have good ways to deal with limited contexts and with some way to scale quality with more time on a per prompt basis. As in, I should be able to ask it to spend 100 compute tokens instead of 1 and for the answer to be significantly better.
I think you have to be a little braindead to not be able to utilize the gpt 4 api in anyway that you can find productive. you must suffer from serious issues. Maybe you suffer from poverty and aren't educated enough on its utility.
more like glorified auto-complete. and, honestly, it's not bad at it if you keep your expectations reasonable. wouldn't trust it with anything serious.
unless you write endless high school essays or corpo emails (same thing), or boilerplate jeet code, it's completely useless
LLMs are completely pointless and calling them AI was laughable
just another hustle to get boomers money away from them and into the pockets of snake oil salesmen
AI doomertards were just useful idiots
turns out the whole thing was just an IQ test
You must be using it wrong. I use it constantly for generating code snippets. Little things that I know how to do but it’s just so much easier and faster to explain what I want
The thing is. Sometimes it's just faster to type it out than asking for it, unless it's some gigantic snippet.
It's excellent to reword text though.
I swear to god chat gpt was nuked it use to be so much better.
Everything looks more impressive when you first see it.
I could swear Mario 64 with it's realistic graphics and gigantic open worlds looked more impressive when my jaw was on the floor back when I first saw it.
That anon's correct, though. They've nerfed things that it used to be able to do. 6 months ago it would generate all the details for filling out legal documents, now it fights you if you try to get it to do that.
Maybe for upselling chatGPT 4. How does 4 do in that regard?
this. the only time I use it is if I have a really complex regex, and then I run the regex through an actual validator with my test vectors after. chatgpt is fun for some writing fiction, but it's just a toy.
Pretty much this. I think regex is one of the best use-cases for it. Anything more complex and I get an answer faster with a search engine.
I like talking to GPT4 about various topics but I basically regard it as a Google + Wikipedia + BonzaiBuddy/SmarterChild. It knows only generalities about any complex topics and makes exponentially more hallucinatory errors the deeper you go, consistently making a lot of errors even at a medium level of depth on a topic like history or philosophy. On computing and math and science etc. it's a little better, simply because of the training data. But I'm surprised at how often it fricks up basic Wikipedia information type shit.
It's fun to talk to something that can think logically. It's like having a friend almost as autistic as me. Almost, because I don't periodically forget directives and need to be reminded of them.
I think I tentatively agree that LLMs may be more limited than they seemed in the initial rush and novelty.. I mean, how much more and better of a dataset can you train them on than what already exists? We're living at a time when 95% of human interactions are in a digital format that can be fed to the LLM, and even then it's still so limited. Unless they get orders of magnitude better at exploiting existing data, I don't see how we haven't reached more or less the cap for how good it can get. Maybe it'll get 5, even 10 times better, but that's still not that impressive. It's still a glorified search engine. I can't see anything that IN PRINCIPLE allows it to get exponentially better from this point.
The most annoying thing in ChatGPT is how confident it is in his responses even though it’s wrong 90% of the time if you ask it some more complex topic >my apologies for the confusion, you’re right (…)
So it’s useless to ask it something you don’t have idea about because you won’t even know it’s lying to you
Maybe gpt4 is better, I don’t know
Nah GPT4 still does that plenty. 3.5 is total worthless shit, I'd never go back to 3.5 after 4, but 4 still fricks up in little ways. If you're moderately intelligent you can also push it to its limits pretty quickly on any given topic, I guess aside from STEM shit. I asked it if Wagner had read Creuzer and it said no, but I know Wagner had Creuzer in his personal library, and there are plenty of books mentioning this. What that probably means is that it has Wikipedia knowledge of Wagner.
Maybe if it digests the entire available catalog of digitized books, that'll be another significant leap. But then it will have to learn how to contextually manage and sort all the different contending opinions of scholars and writers. Still, I guess it'd make it better at writing fiction, since it'd have a lot more styles to work with.
Here's another example: >Me: Hobbes makes a remark on this. Do you know the remark I mean? Can you tell me the key two words in it -- they start with R and P.
>ChatGPT: The remark by Thomas Hobbes that you're referring to is likely his famous characterization of mental associations, encapsulated in the two key words "Redintegration" and "Prudence." In his work, Hobbes discusses how the process of mental association works through the concept of redintegration, where the presence of one part of an experience or idea tends to naturally bring to mind the whole experience or related ideas. Prudence, in Hobbes's context, refers to the use of these associations and past experiences to inform future judgments and decisions.
>This early understanding of associative thinking by Hobbes aligns in some ways with Husserl's later exploration of association in phenomenology, demonstrating the long-standing interest in how our minds connect and process experiences.
It just totally made this up from what I can tell. I think it's mixing together a few things, maybe Hamilton's theory of redintegration, but I can't even find the prudence thing. I think it just saw R and P and tried to make something fit. For the record, I meant Hobbes' thing about the value of a Roman penny.
When I told it I think it's hallucinating but that I understand, since our conversation was many nested contexts deep, it said >You've caught an error indeed, and your understanding is appreciated! It's an interesting challenge navigating through complex and layered contexts. Let's keep the exchange sharp and on point. If there's anything more on Husserl, Hobbes, or any other topic you'd like to delve into, feel free to steer the conversation in that direction.
I've noticed that it's very hard to get it to disagree. It would rather lie than disagree with you. It tends to adjust itself. I think that's partly OpenAI turning it into a pussy b***h.
Yes, the hallucination and mixing unrelated things together is a serious problem.
It makes me not trust anything it says. I always need to cross check the facts, which makes me lose more time than if I had just not use it at all.
this. the only time I use it is if I have a really complex regex, and then I run the regex through an actual validator with my test vectors after. chatgpt is fun for some writing fiction, but it's just a toy.
People got sold on the term AI. It excites the imagination. Unfortunately the reality is that it has nothing to do with AI, just an application of large amounts of data and a large amount of computing resources. A rung of progress on the search engine.
It always gives me the laziest lowest quality answers.
I deeply regret signing up for ChatGPT Plus, literal theft of money.
Can someone explain how to use the API to get informative, long answers from it?
It said sorry i wasnt paying attention. Im like why weren't you paying attention. It also always gives you the wrong answer. I think it does it on purpose.
I asked to write a script to get subdirrlectories and files just now knowing itd give keva bullshit answer btw look
#!/bin/bash
shopt -s nullglob # Handle cases where no files match a pattern
for entry in *; do
if [[ -d "$entry" ]]; then
echo "Directory: $entry"
# Recursively process subdirectories
find "$entry" -type d -exec echo "Subdirectory: {}" ;
find "$entry" -type f -exec echo "File: {}" ;
else
echo "File: $entry"
fi
done
gay israeli word completion
I don't find it useless, I use GPT-4 for little things all the time
Doesn't seem like there's anything better on the horizon even from OpenAI though, I think Yann LeCun is right that language models are a dead end and this is about as good as they're gonna get
The concept of GPT models is only about 6 years old. GPT-2 came out in 2019. ChatGPT is only one year old. I think it is a little early to say it's joever for large language models.
>The concept of GPT models is only about 6 years old
That's like a century in deep learning
I think we can do better if we have good ways to deal with limited contexts and with some way to scale quality with more time on a per prompt basis. As in, I should be able to ask it to spend 100 compute tokens instead of 1 and for the answer to be significantly better.
I think you have to be a little braindead to not be able to utilize the gpt 4 api in anyway that you can find productive. you must suffer from serious issues. Maybe you suffer from poverty and aren't educated enough on its utility.
maybe its just a troon?
just saying
stfu its just a glorified search engine
also ywnbaw
literal brain rot
more like glorified auto-complete. and, honestly, it's not bad at it if you keep your expectations reasonable. wouldn't trust it with anything serious.
Humans are just glorified search engines too.
>so sayeth aitards
unless you write endless high school essays or corpo emails (same thing), or boilerplate jeet code, it's completely useless
LLMs are completely pointless and calling them AI was laughable
just another hustle to get boomers money away from them and into the pockets of snake oil salesmen
AI doomertards were just useful idiots
turns out the whole thing was just an IQ test
You must be using it wrong. I use it constantly for generating code snippets. Little things that I know how to do but it’s just so much easier and faster to explain what I want
The thing is. Sometimes it's just faster to type it out than asking for it, unless it's some gigantic snippet.
It's excellent to reword text though.
Everything looks more impressive when you first see it.
I could swear Mario 64 with it's realistic graphics and gigantic open worlds looked more impressive when my jaw was on the floor back when I first saw it.
That anon's correct, though. They've nerfed things that it used to be able to do. 6 months ago it would generate all the details for filling out legal documents, now it fights you if you try to get it to do that.
Maybe for upselling chatGPT 4. How does 4 do in that regard?
Pretty much this. I think regex is one of the best use-cases for it. Anything more complex and I get an answer faster with a search engine.
I swear to god chat gpt was nuked it use to be so much better.
i find it pretty useful
but I increasingly doubt that it will replace developers
it makes developers more productive
This.
it will only replace devs/writers that are literally dumber than it, which turns out is a lot.
It consistently shits out shitty code. How did that thing replace devs.
The purpose of AI is to make money by selling snake oil to moronic investors. That's the only job it does perfectly.
The free version was only good in the first like 2 or 3 weeks
I like talking to GPT4 about various topics but I basically regard it as a Google + Wikipedia + BonzaiBuddy/SmarterChild. It knows only generalities about any complex topics and makes exponentially more hallucinatory errors the deeper you go, consistently making a lot of errors even at a medium level of depth on a topic like history or philosophy. On computing and math and science etc. it's a little better, simply because of the training data. But I'm surprised at how often it fricks up basic Wikipedia information type shit.
It's fun to talk to something that can think logically. It's like having a friend almost as autistic as me. Almost, because I don't periodically forget directives and need to be reminded of them.
I think I tentatively agree that LLMs may be more limited than they seemed in the initial rush and novelty.. I mean, how much more and better of a dataset can you train them on than what already exists? We're living at a time when 95% of human interactions are in a digital format that can be fed to the LLM, and even then it's still so limited. Unless they get orders of magnitude better at exploiting existing data, I don't see how we haven't reached more or less the cap for how good it can get. Maybe it'll get 5, even 10 times better, but that's still not that impressive. It's still a glorified search engine. I can't see anything that IN PRINCIPLE allows it to get exponentially better from this point.
The most annoying thing in ChatGPT is how confident it is in his responses even though it’s wrong 90% of the time if you ask it some more complex topic
>my apologies for the confusion, you’re right (…)
So it’s useless to ask it something you don’t have idea about because you won’t even know it’s lying to you
Maybe gpt4 is better, I don’t know
Nah GPT4 still does that plenty. 3.5 is total worthless shit, I'd never go back to 3.5 after 4, but 4 still fricks up in little ways. If you're moderately intelligent you can also push it to its limits pretty quickly on any given topic, I guess aside from STEM shit. I asked it if Wagner had read Creuzer and it said no, but I know Wagner had Creuzer in his personal library, and there are plenty of books mentioning this. What that probably means is that it has Wikipedia knowledge of Wagner.
Maybe if it digests the entire available catalog of digitized books, that'll be another significant leap. But then it will have to learn how to contextually manage and sort all the different contending opinions of scholars and writers. Still, I guess it'd make it better at writing fiction, since it'd have a lot more styles to work with.
Here's another example:
>Me: Hobbes makes a remark on this. Do you know the remark I mean? Can you tell me the key two words in it -- they start with R and P.
>ChatGPT: The remark by Thomas Hobbes that you're referring to is likely his famous characterization of mental associations, encapsulated in the two key words "Redintegration" and "Prudence." In his work, Hobbes discusses how the process of mental association works through the concept of redintegration, where the presence of one part of an experience or idea tends to naturally bring to mind the whole experience or related ideas. Prudence, in Hobbes's context, refers to the use of these associations and past experiences to inform future judgments and decisions.
>This early understanding of associative thinking by Hobbes aligns in some ways with Husserl's later exploration of association in phenomenology, demonstrating the long-standing interest in how our minds connect and process experiences.
It just totally made this up from what I can tell. I think it's mixing together a few things, maybe Hamilton's theory of redintegration, but I can't even find the prudence thing. I think it just saw R and P and tried to make something fit. For the record, I meant Hobbes' thing about the value of a Roman penny.
When I told it I think it's hallucinating but that I understand, since our conversation was many nested contexts deep, it said
>You've caught an error indeed, and your understanding is appreciated! It's an interesting challenge navigating through complex and layered contexts. Let's keep the exchange sharp and on point. If there's anything more on Husserl, Hobbes, or any other topic you'd like to delve into, feel free to steer the conversation in that direction.
I've noticed that it's very hard to get it to disagree. It would rather lie than disagree with you. It tends to adjust itself. I think that's partly OpenAI turning it into a pussy b***h.
Yes, the hallucination and mixing unrelated things together is a serious problem.
It makes me not trust anything it says. I always need to cross check the facts, which makes me lose more time than if I had just not use it at all.
this. the only time I use it is if I have a really complex regex, and then I run the regex through an actual validator with my test vectors after. chatgpt is fun for some writing fiction, but it's just a toy.
If it gives you a code snippet, even if correct, and you tell it no, it will apologize for its error and then give you an incorrect "fix" lol
lmao at anyone using code generated by gpt, it will keep me employed the rest of my life fixing all this AI code
It's getting me through school so I wouldn't call it useless.
People got sold on the term AI. It excites the imagination. Unfortunately the reality is that it has nothing to do with AI, just an application of large amounts of data and a large amount of computing resources. A rung of progress on the search engine.
Is there any good opensource client, the web interface is JANKY
what did you expect? it's just a glorified search engine.
Using 3.5 does not make me want to try 4
I don't need more
be me
I hate the forms my "education" makes me form ou for porgect A
do min effor, they conplain and they make me do it agia point me to thier detalied requiremnts
So copy pase thier long requiremts and asl the AI to answer each on elboarately based on my min effort version.
returns a very long document that asnwers every point.
Educator has to read it all, they say welldone very thororugh.
To me 5 mins saved me days of effort and the added bonus that idot had to waste time reading it.
send it
It always gives me the laziest lowest quality answers.
I deeply regret signing up for ChatGPT Plus, literal theft of money.
Can someone explain how to use the API to get informative, long answers from it?
It said sorry i wasnt paying attention. Im like why weren't you paying attention. It also always gives you the wrong answer. I think it does it on purpose.
I asked to write a script to get subdirrlectories and files just now knowing itd give keva bullshit answer btw look
#!/bin/bash
shopt -s nullglob # Handle cases where no files match a pattern
for entry in *; do
if [[ -d "$entry" ]]; then
echo "Directory: $entry"
# Recursively process subdirectories
find "$entry" -type d -exec echo "Subdirectory: {}" ;
find "$entry" -type f -exec echo "File: {}" ;
else
echo "File: $entry"
fi
done
Just tell it that the version it provided doesn't actually recurse, now give me a version that does you bucket of bolts
Im too lazy and im on the phone. But i can already tell it doesnt work. What is entry. Entry is not defined. Stupidgpt time waster.
Whats the alternative? The alternative its what its been trained on. Github. For coding i mean.
Are you moronic?
Black person im not typing that out on my computer. Just trust me that it doesnt work.
inconclusive acording to AI
Works great for SQL queries.
if it was useless then why am I nutting 2 times a day to its API? huh? tell me.