APIs are different from public website access. So, I would imagine that those people who are relying on the API for use in the apps they’ve built on these models would still exist. But I didn’t know how or if this would apply to DTP, so I asked.
Good to know, thanks.
I concur with the other lawyers in this thread. I would much prefer immediate access to the latest OpenAI models rather than a curated selection. I carefully curate and test the models myself to determine if the results are an improvement.
Additionally, I already have ChatGPT Pro, which provides essentially unlimited usage of the ChatGPT web interface. Therefore, if there is a way to utilize that interface in DTP instead of paying extra for API usage, I would strongly prefer that option. I suspect there is a way to configure DTP to achieve this, but the details of the various scripts and their configurations are not transparent.
The question isn’t whether you think the models work effectively, it’s if they work and can be controlled in DEVONthink with privacy and data safety in mind. It is not a trivial matter to support AI models and DEVONthink is not intended to be an uncontrolled hub of AI connections.
1 Like
I understand your concern about exposing an entire DT database to an untested AI.
To narrow the discussion, I’d like to focus on a single document workflow. Specifically, I want to import a scanned document into DTP using ScanSnap and then have DTP leverage my ChatGPT web account (not the API, which incurs additional costs) to analyze the document and propose a filename and tags. Using the web interface is preferable in my case because I have already granted OpenAI certain memories, and I would like to use that existing context to name, tag, and potentially summarize my documents. For purposes of this discussion, please assume that these documents are either non‑private or appropriately sandboxed at OpenAI, so confidentiality is not a concern. My question is: how can I implement this, and why would that pose any risk to my database?
I also understand your hesitation about encouraging broad use of the ChatGPT web interface in this way, particularly if there is a concern that OpenAI might object. However, OpenAI already throttles usage based on account level, and some of us have substantial unused capacity under our existing Pro licenses. I do not believe this usage would exceed my license terms. Functionally, I could manually upload each file to ChatGPT, have it perform the analysis, and then copy the results back. The real challenge is automating the return of that analysis into DTP—specifically, mapping and somehow importing ChatGPT’s output into DTP’s metadata fields in a usable format. Since I am not a programmer or scriptwriter, my assumption is that ChatGPT could help generate the necessary script, if I knew how to frame a sufficiently accurate and comprehensive prompt.
Thanks for all your hard work, I (and I think all of us here) really do appreciate it.
1 Like
then have DTP leverage my ChatGPT web account (not the API, which incurs additional costs)
The API is what is offered to third-parties such as ourselves. In fact, these companies want more people using their API instead of their web interfaces.
Since I am not a programmer or scriptwriter, my assumption is that ChatGPT could help generate the necessary script, if I knew how to frame a sufficiently accurate and comprehensive prompt.
If you’re referring to AppleScript, ChatGPT (and most AI) produce overly complicated and often broken scripts.
Thanks for all your hard work, I (and I think all of us here) really do appreciate it.
Thanks belong to @cgrunenberg, the architect of all this integration and much more.
2 Likes