That is exactly close to my observation as well, I feel truly bad here, specially because the model released to us looks to make lots of errors, it fixes them but it has to apologize and that will count against your cap, even if you supress that but you need it to output something back between the calls that’s where the problem really is and in a flash you are out of cap, waiting.
I love GPT-4 and OpenAI but the system currently doesn’t work, I stopped further development on my plugin last weekend because I no longer able to sustain this.
Bing is my primary use and can’t wait for them to release plugins.
I subscribed to to Plus on the first hour I got the email on the first day they opened up, and really it’s kind of buying couple of devs a cup of coffee for their good work, but the trajectory has been seriously flawed since then.
Respect:
But the idea to charge more or charge per API is actually ridiculous and may not generate them that much money.
OpenAI need to get this to the hand of masses in order to have really a transformation in people’s lives and also make money in the process, not only the few who can afford it and hurdle their expansion and the technology adoption.
20$ is a big bucks for so many people (think students and other people in need) , and it’s actually kind of the higher side, how much you pay for Netflix or any other production studio streaming service? not doing 1-1 mapping but I hope you get the idea of where am I going.
Also what we are talking about here is the inference cost only to be clear.
You can run a good version of LLAMA with a decent performance on your desktop machine with zero cost - your electricity bill
Actually they have to reduce the price and gather as much mass as possible, to make good amount of money in the end.
They can also charge Enterprises which I think would be a huge revenue source and can afford higher prices.
But normal people, students, retiree, etc… that want to use it for their day to day lives should actually be able to get it.
This technology is transformational and the key is to get it to the hand of masses IMHO.
I really want to see OpenAI succeed, they are the first guys that sparked the spark of AGI at the end.
Still love them, but I’m slowly moving from “I can’t live with without this technology”, to what the heck I’m wasting time here kind of thing
Hopefully they listen and I wish them well nonetheless.
I agree completely.
(and yes, I run wizard-13b (a derivative of LLaMA) locally - it’s great!)
However, it is my understanding as a plugin developer that the current situation is that if a plugin, for example my search plugin llmsearch, internally does a callback to the OpenAI api, which mine does, that callback is NOT currently free, but rather is charged to my api account, just like any other OpenAI api call.
Correct , also if you have to prompt it to “Continue”. But generally speaking if it can find the solution without user intervention it will only count as one. It’s depends on the plugin but every time you have a back and forward with it this occurs.
Without even me prompting it .
Think of this as a kind of 4 steps approach where I ask it to display specific output based on the result of the previous step. This will definitely count.
I’m certain because basically it hit the cap limits while executing those steps without me prompting it of anything further. The chain of response it output within that single prompt is definitely counted.
I wish we talk about the great use cases instead of us counting outputs and responses here
Agreed, the 25 message limit will eventually go away. For right now for most cases it is enough. With the 25 message limit the possibilities still seems to be endless.
I was thinking the same, Webpilot scanned about 20 folders from GitHub and my heart sank to my toes. I was able to keep working so for sure, I believe it tallied as a single message which is confusing - I’m happy though.
Unfortunately, organizational access to plugin development is not the default. I still cannot install unverified plugins on the new ChatGPT Plus account made within my organization.
Yes, you can change the manifest file and then go through the “Developer your own plugin” flow again. This will automatically remove your plugin from the plugin store if it has been approved.
I was thinking the same, Webpilot scanned about 20 folders from GitHub and my heart sank to my toes. I was able to keep working so for sure, I believe it tallied as a single message which is confusing - I’m happy though.
