I find those who are using the API will get way better quality, and those who are using the chat get worst quality, yet if using the playground and 3.5 turbo vs 4 I find a lot of time I prefer the content produced by 3.5 turbo vs 4. Yet even with the free version I get great results. Yet I prime the model for what I need it to do. OFC not using functions etc.
Maybe I just understand how to use the free version to get paid quality results.
::shrugs::
I rather train my own models, and make niche LLM's vs OpenAi I do not think the way they train it is going to work and kill it.
Workers caught training AI's with Ai
This is not good and will mess models up...
With less training, smaller niche models do work better then GPT 4.
especially if you teach it on textbooks and factual data only, and not the Whole wide web.
If you are working with these company I wouldnt be using OpenAi directly I would be using a service that keeps your data private... or invest 100k into 2 pcs to train your own models and build out our own API.
How to stop ChatGPT from using your private data
You’ll need to log into your ChatGPT account in a browser to change the privacy settings. After that, follow these simple steps to opt out of ChatGPT data sharing. You’ll notice that OpenAI tries its best to hide the privacy-enhancing feature.
- Click your account in the bottom left corner
- Click Settings
- Click on the Show menu next to Data Controls
- Toggle off the Chat History & Training
Then again I am just a cook, who has been doing this since the 90's as a hobbie. Just a sigma male out of work looking to past time and share my knowledge. So sad seeing these fake GURU's on youtube milking people for income.
@frozen_yoda maybe you can explain how I can make an app script in google documents use Oauth to check stripe for subscription? Dm me if you would like to assist someone.
It's always good to be a hobby cook. That's how I started.
Yes, via the API. I don't use the playground or chat via the browser lol..
Maybe I just understand how to use the free version to get paid quality results.
::shrugs::
This is why I said using the
same engine and prompt.
Using the
same prompts over thousands of calls has produced lower quality on the free accounts vs paid. No shrugs, but actual results.
Workers caught training AI's with Ai
This is not good and will mess models up...
With less training, smaller niche models do work better then GPT 4.
especially if you teach it on textbooks and factual data only, and not the Whole wide web.
If you are working with these company I wouldnt be using OpenAi directly I would be using a service that keeps your data private... or invest 100k into 2 pcs to train your own models and build out our own API.
Perhaps you missed my post from before yours.
Yes, we use our own trained models for confidential data. Not just ChatGPT as I mentioned previously.
I just use the API and can continuously call throughout the day. I've only run into rate limit issues a few times. I usually get access to the new releases same day (it was crazy seeing people wait so long for gpt-4).
Anyone here tried Alpaca** before Stanford redacted some of the code? Worth trying imo. Still use gpt most of the time.
Note, be careful if using gpt-4 with the API. The costs can add up fast.
Edit: Forgot to mention. I have seen "dumbed down" responses using multiple accounts without the paid upgrade. So, I usually just use my plus account.
How to stop ChatGPT from using your private data
You’ll need to log into your ChatGPT account in a browser to change the privacy settings. After that, follow these simple steps to opt out of ChatGPT data sharing. You’ll notice that OpenAI tries its best to hide the privacy-enhancing feature.
- Click your account in the bottom left corner
- Click Settings
- Click on the Show menu next to Data Controls
- Toggle off the Chat History & Training
Don't need your help there. Might be useful to someone else though. I've been using OpenAI for several years and stay up-to-date with best practices.
Then again I am just a cook, who has been doing this since the 90's as a hobbie. Just a sigma male out of work looking to past time and share my knowledge. So sad seeing these fake GURU's on youtube milking people for income.
No one likes the fake guru's or spreading fake knowledge and responses.
-----
I like staying on topic and helping to answer questions people have. Which is what I was trying to do before relating to free vs paid.
----
Going off topic, but I'm happy to provide a high level overview since you asked.
@frozen_yoda maybe you can explain how I can make an app script in google documents use Oauth to check stripe for subscription? Dm me if you would like to assist someone.
I of course can create this and have experience scripting with lots of Google's APIs, including Google Docs. There are lots of ways to do this from an architecture standpoint.
Easiest would be hosting on your own server (or AWS, GCP, Azure) with a subscription based service that has the services integrated including Stripe. I actually have a service setup doing exactly this with multi-tenant setup so they can add the rest of their team. Free, basic, and pro versions for the account with a per user rate. Multi-auth and allow social login using Google for their login or they can setup their own user/pass on the site. I won't be pasting the website here, but uses Node.js for the backend and React.js on the frontend. I use Amazon Aurora, but of course you can use DynamoDB or another db of your choosing. I use lambda functions for calling OpenAI and SQS and poll the queue for responses. Uses cron for various checks since there are multiple APIs and services for user's of my site to check the status of those services and updates related to keywords. This isn't the full architecture nor does it include the full scope of services, but gives a decent high-level overview.
I have been using the latest release for chat functions which has proved be much more efficient than what it was previously when you would request chatgpt to respond with JSON in the prompt. It seemed like 5-10% would result in failure due to not producing a JSON response and I had to add checks before making yet another call for the correct output.
Lots of use cases that are generating $$. For the large company it has been around helping user's get detailed responses from structured data out of various internal and external APIs (SAP, shipping, etc) and for another startup to create customized outreach campaigns using email, LI, and phone. Using a multi-step approach this saves money -> gpt-4 to give high level structure, gpt-3.5-turbo for initial results, and then a loop for final. Costs can add up which is why it's important to look at other LLMs.
Can go on for hours, but might be better off in another thread.
who has been watching AI since A.L.I.C.E and building websites since geocitites.