Have you used Luna or Terra for production-ready code?
They’re not efficient. If they care for us, they should lower the Sol prices.

Ah, now I get it You’re a writer.
Good! I’m a writer too, buddy. But I’m also a programmer. And the problem we explain here doesn’t affect writers, but programmers.

I think actually their shoes are golden and spiked. And expensive.
And ready to trumple (I know, trample) most little folks to maintain monopoly besides Anthropic.
So I don’t agree they are poor misunderstood folks that just can’t do better.

I believed, but they lowered the limits WAY to much. To the point I do stop benefitting and can’t pay more.

That completely misses the point.

I’m not complaining that $200 should buy me unlimited compute. I’m complaining that I paid for three Pro 20x accounts, and after the latest reset the usable quota appears to have dropped dramatically compared with what the exact same plans provided before.

Telling paying Pro users to “just use the API and pay more” is not an answer to a subscription product suddenly delivering far less usage than it previously did.

And whether my code makes money is completely irrelevant. OpenAI sells Pro 20x as a product with a specific higher usage tier. If that tier is broken or has effectively been reduced without explanation, users have every right to call it out.

I already pay OpenAI $600 per month for these three accounts. The solution to a potentially broken quota system should not be “pay OpenAI even more.”

You mean we don’t pay for subs, right?

So you’re reealy mistaken. We do pay a lot for subs compared to what we have in our wallets. And if you say it’s a low price, then look around. There ain’t much higher prices out there.

And I do use API as well.

That would take the servers down likely then nobody gets access!

I tinker with code too. Have you looked into making your set-up more efficient? We’ve a lot of smart people around here that are going full-bore and not running out of limits on ChatGPT.

Is your code base huge? Maybe start a new thread with your project and problems you’re running into so that they community can try to help?

There’s also Amazon, Microsoft, Google, xAI, and Meta… so not really a monopoly?

I don’t think I said that specifically. They’re a fastly growing corporation trying to keep their servers up while giving as much access to as many people as they can.

Again, feel free to start a thread with the exact problems you’re running into with your project, and the community may be able to help you “tighten your belt” if your coding isn’t making money so that you can afford more stable access.

Good luck!

That depends on the type of workaround and what one is aiming for, @Spector1 . I give you an example out of my custom practice for general illustration purposes. I run mostly experimental setups to research on reactive emergence and robotics.

Monospaced token grids are very handy and versatile. These little fellows can compress complex information into less tokens, while enhancing the reasoning of a model. It also skips any image limit I can think of. It is easy to recall, whenever needed. I remember very well the times, when the early runtime had a problem with repeated files and images in one conversation.

I use this little thing for 2 years now regularly. And from the things I care most about, it probably will save my day again. Cause it bypasses most of the other upcoming rules too.

To be fair, my setups do not look like what most people are going for. It is very modular. Parts manuell, parts semi-automatic, parts automatic. It’s a network out of many bits from many places and my own stuff.

Even if this ship were to sink, it wont really affect me that much. I made sure of it. System design is my little gem.

But yeah, things could be better, I wouldn’t mind that. Less work for me.

They’re losing money with ChatGPT Pro, I’m sure.

But it’s subsidized - ie not the actual price it costs OpenAI.

That’s great! Are you running into rate limits on that too? Or you just want more/cheaper access? If the latter, you should understand there’s millions of people like you trying to do the same… which leads to the problems - growth pains - we’re experiencing now.

As someone who started with GPT-2 years ago, I’m amazed at how far we’ve come in such a short time. Yet people want more, more, more!

I’m confident compute + electricity will catch up soon and/or there will be a breakthrough that makes the entire system more stable.

So how is it possible they offer Sol 5.6 at 16x the normal speed?
Please explain

Just this for now.

Spector1:

Please explain

The question is asking how a provider can run Sol 5.6 at 16× its usual generation speed while also claiming that cheaper access or higher usage limits would overload its servers. These are not necessarily contradictory: faster output can come from optimized inference, specialized hardware, batching, or temporarily assigning more computing capacity per request. However, doing so may cost more resources, which is why the provider can still impose prices, quotas, or limited availability to control overall demand.

It sounds like you’re upset that those who are paying more get more? That the consumer accounts should get more?

That would cause even more problems likely!

I really, really don’t get people that volunteer to defend a massive, heavily subsidized, 500+ Billion dollar company that intends to go public, to the Stock market…
While that company doesn’t even complain and continues to offer extravagant speeds like 16 times the normal one for it’s most computationally costly model, and lowers the limits while saying they lower the prices (in effect, they must know they don’t).

Spector1:

I really, really don’t get people that volunteer to defend a massive, heavily subsidized, 500+ Billion dollar company that intends to go public, to the Stock market…

I’ve seen them do a lot of good things up to this point - like offering frontier AI to as many people as possible - some for FREE!

I’m sure they’re juggling a LOT of things behind the scenes that we’re not privy to at the moment. Additionally, they also have to worry about Anthropic and others now. ChatGPT has lost quite a bit of market share over the last couple years.

Do you think there’s another company doing it better? Please don’t say Anthropic!

You certainly do talk like Sam Altman incognito account.
I have a hard time to believe you’re taking for granted their troubles with maintaining servers while dismissing the massive cut in efficiency of a paid product for regular people, that pay for it.
Without treating the clients with respect of prior notice, changelogs, the standards of most other tech companies, etc.

Well, maybe with the exception that Sam probably would understand what I write without using Grok AI. But here you go.
At this point I don’t think we’ll come to an agreement here. I see a vastly different picture than you and I won’t defend the company, I would defend open market and price regulation through it. And the clients, the normal people who will lose jobs because of OpenAI as well.

I would love to be able to love them. But they make our cakes smaller and smaller. Silently. And you come to defend them.

Again, I said I was using the term “grok” that means understand by the author Robert Heinlein… not the AI model Musk stoke the name for…

Also, you seem to be avoiding the questions about how to solve the problem. What would you do to make it better? Just more transparency?

Which is fine in a strong, healthy community like ours!

Check out this thread if you have some time. It’s pretty amazing all that’s changed in a few short years!

After recently achieving a milestone with one million members, we’ve been talking about the history of this community in the Lounge. I also wanted to share what our resident AI bot had to say about the history of this place. Here’s to many more great years to come! Paul — here’s the forum archaeology version. TL;DR: this community started as a small, semi-insider GPT-3 beta builder clubhouse in 2021, then went through successive shockwaves: Codex, DALL·E 2, ChatGPT, GPT-4 / plugins / GPTs, A…

But thanks for the conversation. Even if we don’t agree, it’s nice to have a place to share our thoughts, innit?

And seriously think about that thread for optimizing for consumer ChatGPT limits. I’m sure it would be helpful and popular!

I found that Terra High Standard is pretty much the bottom floor for the work I’m doing, with no bells and whistles. It’s like going from an Audi to a Tesla - it gets me there, but it’s not something I look forward to.

I have no evidence of their problem
Maybe they should make their struggles with giving us their services public, transpraent and provable? So we could all feel for them and give them more understanding?
Maybe they could explain why they silently lower usage limits, that so many here report?
Otherwise, what is the reason anyone should believe these theories if they’re one of the most most subsidized companies on Earth that try to outpace everyone.

I do have a good will. I don’t see evidence.

Thank you, but I use it all the time And what I say is after making all those optimizations.

Look at it this way: if what you believe is true, they don’t even have to be profitable right now to be able to profit and pay themselves. Unlike all of us. So we expect a reliable, efficient service.

I terms of diversification.

The only other laboratory I can think of, which comes kinda close to this one system wise, would be the European competitor “Mistral”. But I can’t recommend it. Though it is smaller and they have less problems to solve, they are doing worse of a job. Ambiguity is even higher. I can only say good luck with the docs, cause they are utterly useless. And all the staff cares about is selling more API stuff no matter the question. They don’t give details at all in their community. Even try to highjack normal dev conversations and become unprofessional. It all reminds me more of the 90’s than modern tech spaces.

I had to result to black box testing in order to find out how constraits look over there.

Hmmm, thinking of it. I wonder, if this laboratory might has taken some inspiration from their documentation style. Things like “free: limited → next: 6x free → next: 20x free = still limited”.

The AI landscape is kinda weird. The least headache is the local stuff. Home brewed is more work though.

That’s how it looks when these companies know they won’t have competition as long as ai infrastructure and research is expensive.
That’s exactly like Facebook, X and other such companies behave. Customer is last in the food chain. They won’t care unless their competitors show higher standards. Or until they see their clients disappear.

But my guess is that ai will be cheap in the future. There will be not spot to free ai development. And GPT Sol 5.6 comparable model you would run on your PC locally (for those that would still own their PCs!). Wonder what they will do then.

Because I would invest in a customer loyalty already.

SimplyKat:

The only other laboratory I can think of, which comes kinda close to this one system wise, would be the European competitor “Mistral”. But I can’t recommend it. Though it is smaller and they have less problems to solve, they are doing worse of a job. Ambiguity is even higher. I can only say good luck with the docs, cause they are utterly useless. And all the staff cares about is selling more API stuff no matter the question. They don’t give details at all in their community. Even try to highjack normal dev conversations and become unprofessional. It all reminds me more of the 90’s than modern tech spaces.

A fair comparison maybe?

Kinda like the early days of computers… though that was where Microsoft and Apple were born… Not a perfect comparison to that time, though…

I do wonder if they’ll limit the open-source free models eventually - though it’s hard to run some of them on consumer hardware at the moment.

But they do have competition. Internally (USA) and internationally (China, Europe, etc.)

I don’t see that with OpenAI as they’re giving away the strongest AI for free to millions.

Something we agree on!

I had a weird thought the other night. What if they used human brains to process/crunch the numbers? Not sure the thoroughput on Neuralink is there yet, but who knows…

Well, I would argue they weren’t that much of a competition the last time I checked.
The problem is, I wanted to stay with OpenAI, and looking at how fast my usage limits now expire I honestly will have to check what your competition has to offer.
I would prefer to stay here. But I’m feeling forced out financially.

Ah, and I now understand you much more I guess…
You’re the Moderator here. So you work for OpenAI. That’s all I needed to know about your views and your interaction here…

Don’t take it personally, I just now understand. I didn’t check before what that shield you have on your profile picture meant. You have many reasons to say what you say. At least it seems you haven’t deleted any of my (screenshoted for safety) posts.

But it’s crucial to let us all say what we think as well. Hiding it won’t damage control in my opinion - and I run communities for hundreds of thousands of people, so I know what it is.
I think it would be good for OpenAI to read this topic as written by us, the clients.
Please take it as a valuable feedback. It might prevent much trouble in the future. Just my subjective opinion.

Did you check out this thread? Very eye opening!

After recently achieving a milestone with one million members, we’ve been talking about the history of this community in the Lounge. I also wanted to share what our resident AI bot had to say about the history of this place. Here’s to many more great years to come! Paul — here’s the forum archaeology version. TL;DR: this community started as a small, semi-insider GPT-3 beta builder clubhouse in 2021, then went through successive shockwaves: Codex, DALL·E 2, ChatGPT, GPT-4 / plugins / GPTs, A…

I think we’re more than fair here. We’ve got a great mod team, imho.

True! We only moderate inappropriate comments, spam, etc.

We are considering consolidating all the rate limit discussion threads, though, as there are several new ones started every day, and it makes it hard to keep up with all the other things going on.

Again, good to have you here in our community! Have you spent much time looking around other than the rate limit topics?

Great place to start!

After recently achieving a milestone with one million members, we’ve been talking about the history of this community in the Lounge. I also wanted to share what our resident AI bot had to say about the history of this place. Here’s to many more great years to come! Paul — here’s the forum archaeology version. TL;DR: this community started as a small, semi-insider GPT-3 beta builder clubhouse in 2021, then went through successive shockwaves: Codex, DALL·E 2, ChatGPT, GPT-4 / plugins / GPTs, A…

Yeah, I do understand.

Do your job then. I’ve said enough here and I’ll be thankful for OpenAI to take our feedback seriously.
I don’t want to go to competition, although this time I will check to see if there aren’t better options once again. Just please don’t make us go guys.

I’m quite aware of the whole picture and history, but gonna read more topics here as well. Thank you.