I noticed that when I move my o1 and 4o threads into a project, because they involve the same student, the o1 tag goes away and it starts acting like a 4o. The only way I can tell this is because o1 doesn’t read PDFs, but my o1 threads do read PDFs when I have them put into a project. Additionally, the o1 mark disappears when an o1 thread is in a project.
I am a Pro user (scraping the money as a school teacher together) because I really want the advantage that o1 offers to analyze the large amount of information about each student to help me make better choices. If putting those threads in a project nullifies o1, then it would be pointless for me to use projects.
Does anyone know about this?
The information in this post is outdated as of Dec 28, 2024. It is now possible to use different models inside projects.
Projects exclusively use 4o.
Currently, there is no way to use another model within a project.
Unfortunately, you cannot chat with o1, or any other models for that matter, inside a project.
I hope this clarifies things for you, and I also hope this will be improved soon.
I think, OpenAI is improving it, and based on users’ feedbacks, they added o1 models into Project Folder, maybe they did not update Help Page information, yet.
This is the thinking process to show it’s O1 (big O? typo?) in a project.
The problem seems to be if you want to include things outside of o1 capabilities in a project. Instructions or files I have noted. I couldn’t access memory either.
You seem to be able to edit the o1 chats ok from inside the project but they seem to be totally independent.
Chats cant see each other so I guess it’s just a foldering capability.
I tested it just now.
The help page is outdated and it’s possible to use different models for project chats.
Selecting 4o allows to chat with PDF files. But the model has to be selected at the start of the conversation.
Here are two sample screenshots from a test project. In the first we are reading the ingredients list from a recipe using 4o.
In the second o1-mini is giving advice on how to improve the recipe.
Note here it is asking o1 a question and o1 knows the project name Test o1. This seems to be all the information it knows outside of the current chat thread.
It is not accessing other information in other chats or files uploaded to other chats.
Similarly in @vb’s example above the second image (o1-mini) is getting information on the Kalter Hund recipe from the llm request and not from any memory or access to the previously uploaded .pdf (4o).
With these well-received updates now in place, I will start working with projects right away.
I expect users will be able to utilize a 4o conversation and extract information from a PDF before creating a new project file using the code interpreter, which can then be referenced in other conversations with different models.
The new project knowledge files will most likely continue to be added manually until o1 is able to handle all required file types.
Store all your project files to avoid ‘hallucinations.’ This way, the AI has multiple references for deduction (similar to the Minesweeper game → where random guessing is reduced because more options/references are available to locate a potential mine. In your case, the ‘mine’ represents the goal or task to be completed successfully).
Make it easier to track and organize project files.
Focus on keeping only successful threads. Adding multiple threads to the same problem with excessive ‘trash’ or unnecessary elements will not improve accuracy.
If you have an o1 model, and you move it into a project, it will act like a 4o model until you remove it. Then it will revert back to an o1 model. That has been my experience.
That is a huge upside. Thank you for sharing that vb. Tell more about being able to reference other chats within a thread? I would love to be able to do that. How does that work?
I don’t expect that’s possible at the moment. Chat conversations are separate entities. Inside a project they are connected via their project instructions and knowledge files.
I would likely go the same route to summarize or distill important information into the knowledge files and access them from another chat inside the project.
I haven’t tried using custom GPTs and actions inside projects, yet. That would open the doors to more practical solutions.
