Not a big improvement relative to the hardware and power it requires. 2 to 3 was a huge leap, 4 is a much smaller leap and can barely run on OpenAIs servers. 5 might genuinely tank the company.
GPT4 is not only slower than 3 and 3.5 while also requiring much more horsepower while not being a big improvement over 3.5. If 5 is an even more marginal improvement over 4 then it might not even be sustainable to keep it running. OpenAI already had to start limiting people using 4 because their servers couldn't handle it, eventually it will reach a point where improvements aren't sustainable relative to what they bring in.
He means he's huffing too much copium.
You can ask 4 to adjust something in a bunch of code and it will automatically adjust everything else that depends on making that change while retaining the rest of the desired functionality. The freaks in aicg that erp with GPT4 can tell you that it 'gets' the big picture of the scene. Like some guy with a fart fetish... His bot asked if it could roll up the car window giving some bs reason and then it farted. So despite doing a single message completion it essentially also predicted the path of the narrative beyond the individual completion.
The ChatGPT version seems kind of gimped in comparison to the API version. It could just be that OAI sucks at writing system messages though
People used the 3.5 version to write whole programs one function at a time.
Garbage in, garbage out.
As an AI language model it's outputs are a reflection of the users inputs. It's important to keep in mind that if you find ChatGPT useless you should refer back to the previous sentence for an explanation. Is there anything else I can help you with?
No they fucking didn't. As soon as it writes more than 200 lines of code it forgets the entire context and starts spewing out garbage
2 months ago
Anonymous
This. Actually, it doesn't get to 200 lines before it starts seriously fucking up. It also forgets things after each query - it drops the previous edits you asked for. It's a colossal waste of time, and no it's not my prompts, it's just shit at coding.
It has a lower chance to hallucinate and has like 25% better prose when writing fiction. This isn't a sci-fi novel it's not getting exponentially better. The real issues will come when people start figuring out how to actually implement it correctly. 3.5 alone has the capability to put huge portions of the population out of work.
The only way that happens this time is if there is a concerted push to create new blue collar jobs, or jobs where interacting with a person rather than a machine (even a capable one) is an important part of the role. Or there'd have to be bullshit/make work jobs.
what new jobs white collar jobs are going to be created that the AI won't just be able to do anyway
>The only way that happens this time is if there is a concerted push to create new blue collar jobs, or jobs where interacting with a person rather than a machine (even a capable one) is an important part of the role.
Based.
FUCK white collar work, it is the most soul crushing thing on the planet. If these corporate losers didn't hoard all the money for themselves nobody would want to sit in a cube for 8 hours a day every day
>new jobs are created due to emerging technologies
Ahh yes, the "we will just create new jobs" hyper copium. For every job that gets created at least a thousand will get replaced by AI and robotics. And those newly created jobs will also get replaced by AI and robotics rather sooner than later.
Might make services that were previously only manageable by large enterprises viable.
For example logistics management on the micro scale.
Why go through some bloated corps outdated system when you can use a much smaller/cheaper service that’s managed by a neighbor? The new income sources will be created by the removal of capital necessary to procure labor since “labor” will be practically free.
>3.5 alone has the capability to put huge portions of the population out of work.
Given how tight the labor market is I'm sure there are plenty of jobs for them to switch to.
AGI is Artificial General Intelligence - the Holy (or unholy, depending on your viewpoint) Grail of AI research. It's the final self-aware, fully functioning replication of human intelligence that is capable of all the same cognitive and creative elements that make up human intelligence. Think "Her" or any of the other AIs you have seen in movies that are capable of fully autonomous reasoning, manifesting a sense of self, and emoting sui generis from autonomous imperatives rather than as a confected pastiche of human behaviours.
There are very important - probably insurmountable - reasons why the creation of genuine AGI is physically and technologically impossible. But it is highly debatable whether we would ever really need such a thing - the improvements in capability on Gen AI are astonishing, and when we get GenAI to assist in its own development, as OAI is already experimenting with, the rate of acceleration of functional ability will rocket. These GenAIs will approach human-type intelligence: they already exceed our abilities in empirical computational terms and speed (e.g. the Chinese Navy now uses GenAI to create the wiring designs for new classes of warship. This is a task that takes a team of naval engineers an entire year, and the Chinese Navy created the wiring schemes for theit latest class of Destroyer in a single day).
So GenAI is approaching the asymptote of what is achievable in empirical computational terms, but qualitatively they exist on a completely separate plane to that on which genuine General Intelligence (of the human type) operates. Nonetheless they will mimic it sufficiently well to pass a Turing Test (which is actually a very primitive and now-obsolete test).
The whole world is saying that a GPT4 codebase leaked to chan and was downloaded thousands of times, but I don't understand how that's possible. What exactly leaked? How is it possible to have installed GPT locally? Is it all horseshit wriotten by journos who don't understand any of this, or was there a portable GPT4 implementation that was available on chan and I just didn;t get here in time?
GPT5 would be probably more focused on data which it works with so they will be more precise and more categorized.
Hard to say if it's going to be game changer or not in this manner but i don't want regulations since they suck and they are crippling AI to write output they would actually give you but they don't because of list.
What will be a breakthrough is AGI if that is somehow truly achieved then it's over not even regulations will help to do anything against it since it would rewrite it self based on truth not based on some trigger words censor list.
4 is a big improvement over 3.5 don't listen to the poorgays ITT
Especially in terms of solving math problems and code challenges, showing that it has a much better raasoning than 3.5
if you listen to sebastian bulbeck's talk he explains that the safety guidelines they've had to put on GPT-4 means it isn't as capable as the base version. So the GPT-4 we are using is nerfed.
With this in mind, I imagine GPT-5, unnerfed, will be close enough to AGI.
The difference between 3 and 4 isn't as big as between 3 and 2 etc. The difference between 5 and 4 will be less than from 3 to 4.
The gays saying otherwise are projecting their cope
I just hope GPT5 can handle exponentially more text and code exponentially better.
I care little about the new "image" and "video" features.
I want to improve its current abilities more than anything.
GPT-4 is much better, but you don't realize it until you play with 3 and then go to 4. The difference is that it just gets things right that 3 doesn't. It isn't hard to find examples of code that gpt-3 produces that isn't quite, right, it gets some minor thing wrong, or doesn't quite do what you need it to, but you try it in 4 and it just works.
4 is still not perfect, though
Yes, because it's smarter. It's like both know what a good text/response looks like, and they try to get that appearance but make it correspond to content that makes sense. Gpt 3 fails at this often, that's why I say you have to try it to see it, because on the surface they seem to behave similarly.
Even if it's just prose, you'll see fewer passages where it is just filling up paper, and more where it surprises you with actual logical content or accuracy.
GPT5 may legit destroy the world.
Not a big improvement relative to the hardware and power it requires. 2 to 3 was a huge leap, 4 is a much smaller leap and can barely run on OpenAIs servers. 5 might genuinely tank the company.
>5 might genuinely tank the company
what do you mean?
GPT4 is not only slower than 3 and 3.5 while also requiring much more horsepower while not being a big improvement over 3.5. If 5 is an even more marginal improvement over 4 then it might not even be sustainable to keep it running. OpenAI already had to start limiting people using 4 because their servers couldn't handle it, eventually it will reach a point where improvements aren't sustainable relative to what they bring in.
He means he's huffing too much copium.
You can ask 4 to adjust something in a bunch of code and it will automatically adjust everything else that depends on making that change while retaining the rest of the desired functionality. The freaks in aicg that erp with GPT4 can tell you that it 'gets' the big picture of the scene. Like some guy with a fart fetish... His bot asked if it could roll up the car window giving some bs reason and then it farted. So despite doing a single message completion it essentially also predicted the path of the narrative beyond the individual completion.
The ChatGPT version seems kind of gimped in comparison to the API version. It could just be that OAI sucks at writing system messages though
>le GPT writes full programs
Yes anon, very cool.
People used the 3.5 version to write whole programs one function at a time.
Garbage in, garbage out.
As an AI language model it's outputs are a reflection of the users inputs. It's important to keep in mind that if you find ChatGPT useless you should refer back to the previous sentence for an explanation. Is there anything else I can help you with?
No they fucking didn't. As soon as it writes more than 200 lines of code it forgets the entire context and starts spewing out garbage
This. Actually, it doesn't get to 200 lines before it starts seriously fucking up. It also forgets things after each query - it drops the previous edits you asked for. It's a colossal waste of time, and no it's not my prompts, it's just shit at coding.
Apologies, I'm a filthy crossboarder
What do you mean by the API version? Is this some way of getting around the built-in content restrictions?
it will kill every person who works there
GPT-5 may legit destroy my dick
Changelog:
- minor fixes and improvements
It has a lower chance to hallucinate and has like 25% better prose when writing fiction. This isn't a sci-fi novel it's not getting exponentially better. The real issues will come when people start figuring out how to actually implement it correctly. 3.5 alone has the capability to put huge portions of the population out of work.
>3.5 alone has the capability to put huge portions of the population out of work.
Good, I want to see the world burn.
>people lose their jobs
>new jobs are created due to emerging technologies
>the world doesn't burn
Sorry anon.
The only way that happens this time is if there is a concerted push to create new blue collar jobs, or jobs where interacting with a person rather than a machine (even a capable one) is an important part of the role. Or there'd have to be bullshit/make work jobs.
what new jobs white collar jobs are going to be created that the AI won't just be able to do anyway
>The only way that happens this time is if there is a concerted push to create new blue collar jobs, or jobs where interacting with a person rather than a machine (even a capable one) is an important part of the role.
Based.
FUCK white collar work, it is the most soul crushing thing on the planet. If these corporate losers didn't hoard all the money for themselves nobody would want to sit in a cube for 8 hours a day every day
>new jobs are created due to emerging technologies
Ahh yes, the "we will just create new jobs" hyper copium. For every job that gets created at least a thousand will get replaced by AI and robotics. And those newly created jobs will also get replaced by AI and robotics rather sooner than later.
>you will have no hope for the future, you shall be outraged and afraid
Anon, go post some anime girls or something.
UBI tyme
Name one real and new job created. I'll wait while I'll enumerate a long list of jobs that will completely disappear in the next three years.
Look guys we have a true believer here.
how will chatgpt do plumbing, electrical work, construction, or even stock shelves?
A sufficiently advanced AI could design robots to perform all those tasks.
as of right now i'm not buying it
Fair enough. Machine learning could just plateau suddenly and then stagnate. We may never get to see what sci-fi horrors await.
>new
????
Might make services that were previously only manageable by large enterprises viable.
For example logistics management on the micro scale.
Why go through some bloated corps outdated system when you can use a much smaller/cheaper service that’s managed by a neighbor? The new income sources will be created by the removal of capital necessary to procure labor since “labor” will be practically free.
>3.5 alone has the capability to put huge portions of the population out of work.
Given how tight the labor market is I'm sure there are plenty of jobs for them to switch to.
5 will be AGI
What's AGI and how is it different than chatgpt 4?
AGI is Artificial General Intelligence - the Holy (or unholy, depending on your viewpoint) Grail of AI research. It's the final self-aware, fully functioning replication of human intelligence that is capable of all the same cognitive and creative elements that make up human intelligence. Think "Her" or any of the other AIs you have seen in movies that are capable of fully autonomous reasoning, manifesting a sense of self, and emoting sui generis from autonomous imperatives rather than as a confected pastiche of human behaviours.
There are very important - probably insurmountable - reasons why the creation of genuine AGI is physically and technologically impossible. But it is highly debatable whether we would ever really need such a thing - the improvements in capability on Gen AI are astonishing, and when we get GenAI to assist in its own development, as OAI is already experimenting with, the rate of acceleration of functional ability will rocket. These GenAIs will approach human-type intelligence: they already exceed our abilities in empirical computational terms and speed (e.g. the Chinese Navy now uses GenAI to create the wiring designs for new classes of warship. This is a task that takes a team of naval engineers an entire year, and the Chinese Navy created the wiring schemes for theit latest class of Destroyer in a single day).
So GenAI is approaching the asymptote of what is achievable in empirical computational terms, but qualitatively they exist on a completely separate plane to that on which genuine General Intelligence (of the human type) operates. Nonetheless they will mimic it sufficiently well to pass a Turing Test (which is actually a very primitive and now-obsolete test).
The whole world is saying that a GPT4 codebase leaked to chan and was downloaded thousands of times, but I don't understand how that's possible. What exactly leaked? How is it possible to have installed GPT locally? Is it all horseshit wriotten by journos who don't understand any of this, or was there a portable GPT4 implementation that was available on chan and I just didn;t get here in time?
Llama leaked. As far as I know GPT4 hasn't leaked.
>GPT-5 is out
GPT-6 is actually going to destroy the world
I hate these retarded stock images
It depends.
GPT5 would be probably more focused on data which it works with so they will be more precise and more categorized.
Hard to say if it's going to be game changer or not in this manner but i don't want regulations since they suck and they are crippling AI to write output they would actually give you but they don't because of list.
What will be a breakthrough is AGI if that is somehow truly achieved then it's over not even regulations will help to do anything against it since it would rewrite it self based on truth not based on some trigger words censor list.
4 is a big improvement over 3.5 don't listen to the poorgays ITT
Especially in terms of solving math problems and code challenges, showing that it has a much better raasoning than 3.5
>reasoning
if you listen to sebastian bulbeck's talk he explains that the safety guidelines they've had to put on GPT-4 means it isn't as capable as the base version. So the GPT-4 we are using is nerfed.
With this in mind, I imagine GPT-5, unnerfed, will be close enough to AGI.
Gpt 4 is already AGI. It's just not ASI.
is there anyway to "unnerf" it?
The difference between 3 and 4 isn't as big as between 3 and 2 etc. The difference between 5 and 4 will be less than from 3 to 4.
The gays saying otherwise are projecting their cope
I just hope GPT5 can handle exponentially more text and code exponentially better.
I care little about the new "image" and "video" features.
I want to improve its current abilities more than anything.
GPT-4 is much better, but you don't realize it until you play with 3 and then go to 4. The difference is that it just gets things right that 3 doesn't. It isn't hard to find examples of code that gpt-3 produces that isn't quite, right, it gets some minor thing wrong, or doesn't quite do what you need it to, but you try it in 4 and it just works.
4 is still not perfect, though
what if you don't code is it still better than 3.5?
Yes, because it's smarter. It's like both know what a good text/response looks like, and they try to get that appearance but make it correspond to content that makes sense. Gpt 3 fails at this often, that's why I say you have to try it to see it, because on the surface they seem to behave similarly.
Even if it's just prose, you'll see fewer passages where it is just filling up paper, and more where it surprises you with actual logical content or accuracy.