With Subscription no GPT-4 engine available

OpenAI api engines.list does not have GPT-4 and calling it, 404 not found. When will this be available to us street dogs :dog2:? I can at least use the 4.0 Chat on the OpenAI site now. I see from the docs same endpoint v1/chat/completions. The playground does not list it either.

1 Like

All the information we know is here:

https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4

From the wording I’d guess in next day or so, but no specifics are available.

1 Like

Makes sense. I just now got my subscription so I am at the end of the line for access based on what I have read. From what I have seen of it so far I don’t see much difference. Nothing there that comes close to competing with Bing ChatGPT. Seems like without being able to train it I am not sure I want to use it anymore. Azure may be the only place I can get a bot that is as least as good as Bing Bot. Web search images real time info live hyperlinks idk :woman_shrugging: did openai sell us out to Microsoft and not keep anything good for us ? Seriously why should I continue trying anymore ?

Azure doesn’t have GPT with web search or images as far as I know.

It’s totally OK to just use Bing Chat or ChatGPT for your everyday life if they work for you. You don’t need to use the API unless you want to flex your programming muscles or (like most people here) want to integrate the AI’s language ability into their business’s processes and applications.

1 Like

You can “train” GPT-4 by using embeddings. Most folks here want to add their knowledge to the model, and assume fine-tuning is how you accomplish this, however embeddings is the direct and most successful route for this.

But, specifically, what are you trying to do with GPT-4 again?

2 Likes

I was toying with the idea of providing pre-trained buisness bots that could be easily deployed and fine tuned by a business. But from what I have seen so far I cannot do it with 3.5 I have some limited success using the assistant and system inputs but the results are not consistent or really business ready.

Being able to call a link for example to read a file simply has to be overlayed this five minutes. Which is okay.

But training a deep template with all the bells and whistles a buisness needs is out of scope at the moment.

I am not aware of embeddings perhaps when I get api access to 4.0 I can try it. I am really trying to stay with OpenAI

But without the real-time searches like Bing has and image creation I am not sure I can be successful doing it and I worked for Microsoft I just cannot give them a dime :sunglasses:.

Idk :woman_shrugging: appreciate the discussion

It sounds like you need to check out “embeddings”. Just search this forum!

But basically, you have all your company facts, you correlate the incoming question with your facts, then feed the facts into the prompt and ask GPT-4 to answer the question with respect to the facts.

To put more meat on this, check out this notebook. Also check out the entire repo.

2 Likes

You might have saved the day thankyou very much. I will indeed dive into the rabbit :rabbit2: hole I will let you know how it went if I make it out :sunglasses:

I did a while ago look at how to do some of this creating a vector database to search but I didn’t go far enough apparently to see its power as a overlay for a target business model.

I will create the pipeline necessary and some sort of data store then as I briefly caught scanning your link I would need to over lay the user input do the vector search for the chunk I feed into 3.5 with the user’s questions

Looks like a good amount of work perhaps for August before i can really make any assessment of my ability to create say a small store or restaurant bot.

The ability to create accounts store credit cards post tickets etc would need to be built in and have the new proud owner the work of customizing it for their own place

2 Likes

Well I can create a .csv with my chunks then search for any word in the csv and get the phrase

I am happy for tonight


Making progress


Not sure how I am going to implement it yet but this 5 min if it doesn’t find anything in the embeddings file it won’t even call GPT

Locking it down really hard this way you can try it and play with the embeddings file if you want on arcticfoxltc.com

This is the answer clearly. You can lock down the bot to only responding to a word it finds in the embeddings file. This has hope of becoming a real trained bot I am going to train this one all about https://talkeetna.info