Wow… Nice! The responses are absolutely amazing
1 Like
I agree - the wording intimated this possibility, but if you’ve studied how these LLMs are built, you would be sceptical, eh? 
I’ve been using exactly this approach for a while now, even before GP4. Using the API, creating an interface that supports this process is relatively trivial, and GP4 makes it far easier to pull it off in one API call.
Given a query…
- Use GPT to extract entities, keywords, and links.
- Blend the key content sections from each referenced link into the prompt
- Off to GP4 and an amazing few-shot result.
1 Like
But, you can give any GPT models the sensation that they do indeed have access to specific Internet content using few-shot steps in sequenced API calls. GP4 may reduce the shots while expanding intelligent results involving today’s Internet content.
Upside…
I actually like the idea that you can build AI applications without dependencies on historical cut-off points. Using multi-shot processes opens the door to factoring in real-time data on the Interwebs with outcomes shaped by the LLM. I’ve also experimented with embedding vectors fabricated with information the LLM has no awareness of, and it works pretty well.
I’m relatively new to embeddings, but they seem very powerful and easily “trained”.
1 Like
Yes, of course we can take current data from the network and feed that information back into an LLM to get results with current data, but that is very different technically, than stating the pre-trained LLMs have Internet access. They do not.
We can also fine-tune with current data, but that is creating a new model and that new model will also not have Internet access. It’s just fine-tuned.
HTH
1 Like
Indeed. If it seems like they do, it’s an illusion. I’ve created a very compelling illusion blending real-time streaming highway analytics with LLMs. I had to deflate the users euphoria in three explanation attempts.
These are very powerful alchemies but they can be visually misleading to unsuspecting users.
2 Likes
Very excited about it, I signed up and paid, I can log in the old interface GTP Plus. then select GTP 4 But nothing really seems to have changed, when I try to insert more than 2000 words and ask for a summary, It just comes back with:The message you submitted was too long, please reload the conversation and submit something shorter.
1 Like
Enhancing GPT-4 API with Metadata and Segmentation for Versatile Use Cases
I think it could be nice to add an extension to the GPT-4 API that would incorporate metadata and segmentation in the replies generated by the language model. This would allow users to easily adapt the responses for various use cases, ultimately increasing the flexibility and usability of the AI model.
One potential implementation is having the AI generate code blocks with metadata. This would enable users to create new files or perform other tasks, as long as the implementation is handled by the API consumer. While this isn’t the only use case I have in mind, I believe it’s a valuable example of the possibilities this extension could offer.
Another interesting application would be the ability to output ANSI color sequences for terminal-like displays. This feature would further expand the ways users can interact with and display the AI’s responses in a visually appealing manner.
In addition to these examples, I envision numerous other use cases that could benefit from the proposed API extension. By enabling the AI model to output sections in specific configurations with metadata, we can augment the capabilities of GPT-4 and allow users to leverage its full potential in a variety of scenarios.
I believe that enhancing the GPT-4 API with metadata and segmentation features would bring immense value to users and developers alike. I am eager to hear your thoughts and ideas about this proposal, and I hope we can collaborate to make this vision a reality.
Sorry if I do not sound like if I wrote this myself as a human person but you know I am an enthusiast user of our beloved ChatGPT+ 3.5 / 4
Hi, This is Seong Choe, Family Nurse Practitioner.
I run an outpatient clinic, working as primary care provider.
I would like to apply my experience, knowledge to ChatGPT for better care of the patients.
My email is;
yeniyani@gmail.com
thanks.
1 Like
Same question. Will we be constrained to the fine-tuned conversation model? Is it for safety purposes?
1 Like
Hello,
Are there plans to add GPT-4 to the playground?
1 Like
It’s under Chat completion tab in Playground for me…
Can someone please tell me how to access the tool to upload a photo into ChatGPt 4 ?
I think I have everything setup, and I log in, and select GPT4, but all I can do is type questions.
What am I missing?
@logankilpatrick It looks like the image ingestion is not activated yet. Is that correct, or am I doing something wrong ? If it is not activated yet, and I am sure, you may not be able to comment, but I thought I’ll give it a try : Any time frame for that ?
If the previous reply applies for me too, how can I activate the ability to upload images?
Thank you for access to the API. Here is what we are building ChatStart - Build ChatGPT powered business. Fast!
- Love that API did not change between 3.5 and 4. We just had to swap the model parameter out to make our app work
- Noticing the difference in certain use cases like multilingual and math reasoning.
- Cannot wait to access multimodal as our use case is already mashing up DALL.E and ChatGPT in same session and chatbot UX.
You guys are awesome!! Keep shipping fast…
I kindly advise everyone to read this GPT-4 research report by OpenAI, dated March 15, 2023 (hot off the press), some of which I have summarized in excepts for those who don’t like (or have the time) reading research papers;
2 Likes
Hello, the GPT-4 API registration keeps saying that there is a problem with my application over, and over again.
@logankilpatrick Yea, that’s nice. Would be even more nice if chat history came back since I lost all my progress with GPT-4. It’s really hard to work on anything if the progress is just lost in one moment… 
Je suis vraiment satisfait quand cela fonctionne et cela fait 2 jours que rien ne fonctionnent pas d’historique et je suis obligé de recommencé et quelques minutes apres rien ne s’affichent et il faut tout recommencer… .je laisser plusieurs messages . Comment faire pour avoir la solution…