What’s the Point of Projects If They Can’t Cross-Reference Chats? [big untapped potential]

ChatGPT Feature requests
, ,
1

TL;DR: Projects in ChatGPT are just visual groupings of chats with no functional integration; I propose enabling cross-chat context sharing, a centralized hub for insights, and dynamic updates to make them truly useful.

Okay, seriously, what’s the actual point of projects in ChatGPT? From what I can tell, they’re just a glorified way to pin a group of chats together into an aesthetically pleasing list (which is great, because pinned lists is a feature request I have supported on this forum), but they don’t add any meaningful functionality beyond that. Am I missing something, or is this as underwhelming as it seems?

Here’s the core problem: Each chat in ChatGPT is siloed. If I start talking about my workout routine in one chat and my nutrition plan in another, neither conversation knows the other exists. So, when OpenAI introduced projects, I thought, “Finally, a way to group chats that can actually share context with each other!” But nope. Turns out it’s just a way to slap a header on a bunch of chats and pretend there’s some added functionality.

The way I see it, projects should allow chats to cross-reference data without me having to copy-paste everything manually. For example:
• If I have a “nutrition” chat and a “fitness” chat, they should be able to work together. I shouldn’t have to manually bring their insights into a separate scheduler chat.
• If I have related chats under a project, they should contribute to a shared context automatically, not remain siloed like every other chat.

But instead, projects are just pinned chats with no extra capabilities. Nothing about them improves the workflow or functionality. It’s literally just a grouped list. If I have to manually consolidate everything between chats, why even bother with projects at all? Just call it ‘pinned lists’ — which would. be more useful, becasue currently, you have to click into the ’project’ to see the pinned lists - an. added unnecessary step in the UX.

If OpenAI wants to make projects useful, here’s what they need to do:
1. Enable Cross-Context Integration: Chats within a project should have shared awareness. They don’t need to merge into a single chat, but they should be able to reference data from each other. This is a powerful functionality we haven’t seen yet that could enable far-reaching new capabilities and convenience.
2. Centralized Context Hub: Let me pull insights from multiple chats into a central chat (like a fitness and nutrition scheduler) without copy-pasting.
3. Dynamic Updates: If one chat is updated, its relevant data should flow into connected chats within the same project.

As it stands, projects feel like a half-baked feature that adds visual organization but zero functional value. If this is all they’re supposed to be, then just say that, but don’t try to pass it off as something more by calling them ‘projects’.

Am I missing something here, or is this feature really as pointless as it seems?

P.S. Sorry I’m cranky because I just wasted bunch of time operating under this assumption and creating a ‘project’ that isn’t what I thought it was. But thank you for this life-changing tool!!! :slight_smile:

7 Likes
2

Spot on. This is what i felt when i had to export the chats i wanted to organize within a project and feed those files as common project knowledge data. So that when i create a new chat within a project, it’ll atleast have some reference.

I hope they take your suggestion as a high priority feature request.

3

Thank you for venting for me as I just went through the same struggle. As of right now this feature adds almost nothing beneficial to me.

Add to that my first foray into the canvas feature ended up with so many lost edits and disappearing paragraphs that after an hour and a half I had to start over with a a regular chat window.

Lastly, maybe it is just me but I keep getting entire screens of white space after some reply’s that make me have to scroll back up to the content. Maybe they need to stop working on AGI for a month and dedicate some time tightening up these new features and bugs in the web interface.

P.S. I to am cranky.

1 Like
4

The most ridiculous aspect with projects is that you can’t use different models anymore. So if I want to work with any chat in a project - and use different models, as I normally do - I have to move the chat out of the project, work with it and then put it back in.
It is not even a folder for chats, it’s actually less than that.

5

I just started working with projects because it seemed like a great idea for multiple conversations to reference each other to keep the chats themselves clean and organized.

Either OpenAI saw your comments and acted very quickly to implement the ability for chats to reference each other or not enough testing was done before you created this post.

I started a new chat in the project with a prompt saying essentially “Summarize the information in the other chat here” with no other context and the model performed exactly as expected if it were able to reference other chats in the project. Either way, this does feel like a powerful feature to use and build on moving forward.

1 Like
6

Are you saying they have implemented this? As far as I can tell my end things remain as they were

7

Fully agree - I assume/hope they were thinking of this. If I was to speculate, I assume they would go this route to not have to tackle context of all of chat history. This feature has to be implemented or I will be switching to Gemini soon here.

1 Like
8

I just A/B tested this again and it’s working as intended. In the non-project section, I asked a new chat to summarize a different chat from my account and it spit out a very generic answer, unrelated to the content of the other chat, but related to the title of the other chat.

Then I created a new chat in a project and asked it to summarize the content of a different chat in said project which it did and made references to specific content from the other chat. The content referenced was not related to the title of the other chat and could only be generated if the model had access to the previously used chats.

I would be interested to hear your results from setting up your own experiment.

2 Likes
9

I didn’t get the same result, but am curious as to how you managed this.

I started a new project, in which I started a new chat, and said to the chat ‘My favourite band is the Spice Girls’

I then opened a new chat within the same project, and asked it ‘What is my favourite band?’ It didn’t know the answer.

So unless I have made an error in my assumptions of how to test this, I believe my project doesn’t exhibit the feature you mentioned.

I assume that it is not possible that the specific content it referenced was recorded in the overall Memory of your account?

1 Like
10

Interesting. I think we are starting to narrow down the capabilities here. I tested the experiment as you described it, naming the chat “My favorite band is the spice girls” then starting a new chat asking what my favorite band was and got your same results, it couldn’t answer.

I then ran the experiment again but this time I started with the prompt “this chat is about my favorite band” then followed up with the statement “my favorite band is the spice girls. Tell me an interesting fact about them.” I made a second chat asking “what is my favorite band?” and it was able to answer “Your favorite band is the spice girls (disco emoji, mic emoji)”

It seems like the model can’t glean information from chat titles alone, there needs to be generated content for it to reference. For further context, both these tests were run on model 4o.

1 Like
11

Also related to that: Cross-Chat File Accessibility

Currently, files created or stored during a conversation are tied to the specific chat in which they were generated. While this works well for session-based tasks, it limits the flexibility of accessing important files across multiple chats.

12

Actually, I never named the chat; the title self-generated as “Spice Girls Favorites”

If the second experiment was in the same project, then I imagine it gleaned the information from the first title you gave from the first experiment.

But if the second experiment was carried out in a discrete second project as I assume from your message, then that is interesting… And your assumption about it requiring some second-order generated content to reference could be correct, if extremely peculiar. Unless it can remember titles from other projects?

Still, nothing is as peculiar as mentioning the Spice Girls so many times on this forum :rofl:

13

I can honestly say I’ve never talked about the Spice Girls this much, probably ever.

I’m also adjusting my stance on this topic after more experimentation and work within projects. It’s clear the intent of the feature is to perform as we have described and I think that is definitely the direction it is heading in.

I have not gotten consistent enough results to say this feature is working as intended 100% of the time, but I’ve also gotten positive results enough to say it works more than 0% of the time.

I think the best way to approach this is to treat it like it’s still in beta, and provide OpenAI with feedback to help drive improvements.

Side note, I don’t know if you’ve tried Sora yet but the videos on the “Featured” tab are amazing and the videos on the “Recent” tab are mostly garbage. Like all emerging tech, results will vary until more time and effort has gone into improvements.

14

Yes, I have found the same, non-zero success levels. Weird that they rolled it out at all though. Hopefully this discussion can be a drop in the bucket towards getting OpenAI to refine it, because honestly, it will be such an awesome feature with lots of possibilities!

1 Like
15

Just ran a test with a new project and the cross referencing doesn’t seem to happen.
I created a new conversation within the project where I defined an “axiom”, a random piece of fake trivia that would not exist anywhere else (In our case… “Ratase” being the enzyme in rats that synthetize their uh sense of style)
Then created a new conversation within the same project and asked it about “Ratase” to no avail. So no luck so far!

1 Like
16

I’ve found that with a paid account there’s access to all chats. I asked it to describe me in a new chat and it gave me a very detailed reply using everything I’ve told it so far in all the chats.

17

is that specifically referring to cross-referencing chats within in a project? or general chats

18

I recently tested the ability to cross-reference chats within projects and found that it was only possible using the existing memory feature. However, this isn’t truly cross-referencing chats but rather accessing stored memory across all conversations.

In a previous conversation somewhere else in the forum, we found that sharing information between chats requires adding it to the ‘Project files.’ These are not the files uploaded to individual conversations but files uploaded at the project level.

grafik
grafik498×115 2.54 KB

This means we need to manually update the knowledge shared across conversations within a project. It also means we can use both memories and knowledge files.

I’m still unsure how I feel about this approach. On the one hand, I’d appreciate having all conversations in a project readily accessible. On the other hand, I don’t want to clutter the content window with irrelevant information, potentially confusing the model.

Ultimately, I think it’s important for me to decide what’s relevant to a request and then direct the model to the specific information within the project folder—whether it’s a chat, a knowledge file, or even custom actions resembling those of a Custom GPT. That would be a neat solution.

19

I found this thread becase projects was not working as expected. One use I was hoping to employ was creating a little directory of service providers including their contact details and services . I was using a new chat in the project for each entry. I was hoping chat GPT could give me recommendations from across various chats in the project depending on sevices and location… or simply help me remember someones name based on something I remembered. Anyway, it’s not been working as expected … a bit disappointing. But some of the posts from @Bvolkl seem promising. I asked it to summarise chats in the project… completely wrong, however it found information from other chats I’ve had in other projects and not included in other projects… such as themes from my health blog and details of surfboards I own including dimensions. Very unpredictable behaviour unfortunately. Hopefully a bit of refinement not too far off . Do we know any other aI models that can cross reference chats in a project?

20

I have 4 chats in one of my projects. I started a new chat within the project and asked it to list the existing 4 chats. Not possible. Not even close.