GPT-4, 40 message every 3 hours now?

[brisk pewter]

GPT-4, 40 message every 3 hours now?

[jagged comet]

Yes, for everyone

[brisk pewter]

GPT-4, 40 message every 3 hours now?

[brisk pewter]

what on earth?!

reduced from 50 to 40? crap

[jagged comet]

It used to be 25, it's adjusted based on server load, and these servers are loaded lemme tell ya hwhat

[brisk pewter]

too bad news

[steady lodge]

This is becoming unforgivable. We need to find another solution, perhaps an open source LLM with GPT4 or close quality. Even being a plus membership, it becomes almost useless. 40 messages at 3 hours, are you kidding ? 😲😲😲

[jagged comet]

If it helps contextualize at all, 40 is one message every 4min30sec, so if you can be careful and thoughtful in crafting your responses and requests to ChatGPT, and give it a lot of detail which is usually what it likes, it can help fill that time and in the end produce a better result for you. But I also want a higher cap, so I feel you there 🙂

[steady lodge]

I understand that you're trying to justify. If ChatGPT had lowered to 10, still someone would come here to justify, and accept. The solution is not to applaud when companies lower the quality of their products and services, but to ask for more resources, in this case, more infrastructure, to guarantee the demand for optimal services delivery to the consumer. Indeed, there should be no Cap at all. We understand that messages should be unlimited for those who has a plus subscription. But since currently this is impossible due to the high demand, we need to keep the advances already won ( not going back ), and providing more servers, more infrastructure from Microsoft to hold all demand costumers need. That is partnership. I hope sincerely that the infrastructure would improve, because the tendency is to even more demand, with the introduction of the new GPTs feature. 🙏👍

[jagged comet]

I disagree with your characterization of what I'd call "coping" as "applauding." And re: "more resources, more infrastructure," I've already seen OpenAI staff acknowledge this as obvious, so I don't think me, as a disappointed user, saying "You guys need more resources and infrastructure" is as insightful or productive as you seem to be suggesting it might be. They've already taken measures like pausing new Plus subscriptions, and really who knows what comes next regarding the surprising leadership changes--considering that tumult, I feel grateful that our only impact as end-users is a 20% slowdown in our potential usage of this awesome beta product.

[steady lodge]

I understand your satisfaction and coping with the latest changes. I will keep asking for more resources. Things get solved not by conformity, but with customers keeping the eyes opened and always focusing on improvements. I hope all problems get solved, and I'm sure there are sensible people to the costumers experience. 👍

[jagged comet]

I don't think customers asking for more solves problems. If anything, it is a problem, for the company, because it means they have dissatisfied customers. Which is totally legitimate! But OpenAI have some nonprofit directives at their core, which is knowledge we have available as customers. https://openai.com/our-structure https://openai.com/charter
This tech contributes to something bigger than traditional economic supply and demand

[acoustic flame]

A question here, how does it refresh?

It refreshes 40 every 3 hours once depleted?, it start once the cap is used (after one prompt) ?, if you deplete it it refreshes only after 3 hours or every 4 min as you said?
So many questions and yet they accuse someone for being "unclear"....Anyways.

[jagged comet]

Think of it like a 3hr timer on each message. You send a message, you wait 3hrs, you get that message back. It rolls like that for all 40 messages.

[acoustic flame]

So a timer that starts and refresh after 3h, up to 40 msg, right?

[acoustic flame]

It wouldn't make sense if you said that each msg takes 3h to refresh and you only can store up to 40 messages

[jagged comet]

What do you mean by "only can store"? I think you get what I'm saying though yes. To recap: you can send up to 40msgs in 3hrs. You get each message "back" after 3hrs, to use again. So think of each message as having its own 3hr timer, and when it runs out, you can get it back. Sorry if I'm being unnecessarily repetitive 😁

[acoustic flame]

No, I mean, if each message has its own timer, imagine using all the messages with 1 min of time span between them, so before 3h you have it all empty, after 3h you only get the first message back? (BC others has 1min of offset)

[jagged comet]

That is correct, if I understand you correctly. If you used 40 messages in 40 minutes, one minute between each, you'd then have to wait 2hrs 20min to get that first message you sent "back." Then, each other message would come "back" over the next 39 minutes.

[acoustic flame]

really thankfull for your explanation, it would be amazing if openai publish it somewhere in case someone like me who are lost.
Edit: as well as if they can implement a way to visualize the timer of the prompts left and sent, but ig it would be so difficult that they will overload the servers xd

[jagged comet]

I like the idea of some kind of visualization! Maybe you could type up a post for #1070006151938314300 on this? Either a timer or an "inventory" of available actions--either or both of those or something else could be helpful!

[zinc basin]

My experience is that how it works is when you send a message, that starts a (roughly) 3 hour timer - once you have used all of your messages you'll have to wait until the end of that but once those 3 hours are over you've got the full 40 back straight away

I've never had a situation where immediately after my messages come back I hit the limit again - in your example this would be quite possible (say you send 1 message a minute from 8.00 to 8.40 - you would get your messages "back" at 11.00 but even if you had sent 5 in that first minute you wouldn't hit the limit again until you'd used all 40)

[acoustic flame]

Umm I would really want a official statement of this issue now XD

[zinc basin]

IMO a counter of actions remaining would be good though - especially as I definitely don't seem to get 40 atm so seeing how the numbers are worked out transparently would be helpful

[zinc basin]

There are times where if I know I want to be able to message on my lunch break, I'll deliberately send a message (for the first time that day) 3 hours and 10 minutes before

[jagged comet]

"I've never had a situation where immediately after my messages come back I hit the limit again"

I've seen this myself--it told me when my next message could be sent, I sent one then, tried to send another, and was given another time to wait to send another, like it was "rolling"

[acoustic flame]

I had this issue using a "glitch", so I got the limit on web version, it said come back at 7:30, got on my phone and kept using gpt-4 (idk why it works) but after 7:30 it told me to wait until 7:35

[jagged comet]

I have also seen this too!

So it's definitely "weird" rn. I think the rolling limit is how it's intended to work though. Something is getting lost in execution for some user experiences though, clearly

[zinc basin]

Tbf I don't mind how it works as long as I know what I'm getting

The very opaque nature of the system (and the fact some people definitely don't seem to get 40) makes it really unclear which isn't great

[acoustic flame]

I mean, before isn't any limit written apart from "dynamic adjusting", now with the clear limit of 40, at least you should be able to spam 40 prompts one after another.

[zinc basin]

No - this is really more for a separate bug report but I definitely only get 15-20 before I hit the limit (including retries, errors within that 15)

[acoustic flame]

Either way, you should be able to not be capped if you send a prompt every 2.22 min

If everything works as intented

[zinc basin]

I imagine they may be testing different limits on different users to see how these affect satisfaction, demand, etc - certainly having a big impact on me as I've gone back to using 3.5 as my default and only 4 quite rarely (which I did when the limit was 25)

[acoustic flame]

Idk men, it fells "weird" to use 3.5 after you payed for the 4.0.
I keep thinking it would give me wrong or dumb answers (I use it for study, as a university student)

[zinc basin]

Yeah but at least I know I'll get something (and the quality is better than I remembered) - 50 inputs nearly always lasted me three hours (or close to it), 15 inputs almost never does even if barely use 4

[acoustic flame]

Is there also cap for 3.5?

[zinc basin]

Not as far as I'm aware no

[steady lodge]

I agree with you. ChatGPT should have a display to inform how many messages can we still post without reaching the CAP. Would it be so difficult to implement ? This would be beneficial to the servers also. 🙏👍

[acoustic jungle]

I hit the limits after 25 in My custom gpts

[grim echo]

Try out ollama platform.

[brisk pewter]

Back to good old days now? 25? Haha. I also just encountered 25 prompt limit on gpt4. Now i think i just sleep and continue tomorrow to prompt more😂

[steady lodge]

Yes, Ollama is a good option. LM Studio is another option. I'm preparing a new machine for running LLMs locally. My computers are very old, that's why I use web based LLMs. But if you see on Hugging face, the newest LLMs are incredible, and we can automate them locally, with other tools like memgpt, retrieval, etc

[acoustic flame]

So? Where did the 40msg/3h gone?

[brisk pewter]

no idea

[brisk pewter]

no cap

[fair pollen]

I'm really not happy with OpenAI. I probably waste half of my 3hr limit message because it gives me an error. If they're going to lower it than for Plus members we shouldn't still be getting charged as much.

[meager musk]

I'm seriously considering canceling my plan. It's giving me a lot of errors, the new usage limit policy is awful, having a separate timer for each message is ridiculous, it makes it impractical to use. Moreover, we're paying to create GPTs that are improving the overall usability of the tool, and so on. I found it terrible. They had already given 50 messages every 3 hours for GPT-4, so why go back and change the time counting policy to restrict messages? Especially when it often gives some error, it's always giving errors all the time, whether in generating images, searching the internet, or analyzing data... it's getting terrible.

[acoustic flame]

im pretty sure you arent the only one considering that xd, i already made my mind with the huge blunder they made with the CEO changes and stuff

[fair pollen]

This is totally me! Lol

https://www.instagram.com/reel/CyhbU-zxIaw/?igshid=MzY1NDJmNzMyNQ==

davedavemicrowave

Likes

336556

Instagram

[meager musk]

now, even if u pay, u dont have acess, ChatGPT just stop to answer and its out

it says its at full capacity, well, we pay to have it at peak hours right ?

[daring oracle]

Very disappointed

[meager musk]

my history has gone, I dont even have access to my history..., it´s the fifth time this week

we need an alternative, I will test Claude 2.1, it´s not the same though I have to find alternatives

it´s in Portuguese, it´s saying that it´s at full capacity

totally agree

[fair pollen]

OpenAI is a dumpster fire right now.

Mine won't work at all today.

I'll be letting my bank know the monthly charge is not valid due to services not rendered.

[brisk pewter]

I think chatgpt service is downed at the moment

https://status.openai.com/

[proven tangle]

its the new ceo

[proven tangle]

like how are we going backwards

[brisk pewter]

Sam is back

hoping for 75prompt/3 hours

[crystal surge]

Lately I spend 30/40 questions just trying to get it to push an answer out instead of some error message 😐

Anyways it took way too long to create just 4 custom GPTs. I had to think ahead about the questions and come up with a template right away for the configurations or theres no way to push anything out in a serious way. Right now this is just a play toy like this and a frustrating $20 a month play toy that works 1/3 of the time 😦
Doing real data analysis or any real coding projects really just sucks right now which is exactly what I'm building some of my GPTs around to help me with for my course work. Not just a play toy to me, this is supposed to help and yet...

[acoustic flame]

Didn't he gone to Microsoft?
Edit: nvm I just saw it XD

[brisk pewter]

Just got back

[brisk pewter]

Maybe draft your prompt in notepad first, then submit

[crystal surge]

So, take all the automation out of an ai? lol At this point its basically CarrierPigeonGPT lol

[brisk pewter]

i totally need the 50 usage caps. i just hit the limit

[cold python]

They really need to bump the server power

[jolly nest]

I hope they will increase the limit to 100/3h or more asap... 40 is really, really annoying... :/

[acoustic monolith]

40 as cap normal for Plus users....really not practical

[grim echo]

I am using this high volume for customer interactions at work. Who do I contact to migrate myself and my coworkers to an enterprise plan?

[jagged comet]

"Contact sales" from here: https://openai.com/enterprise

[daring oracle]

What's the deal 🤔 I can't do anything right now is the service down?