And have you tried another model than Nano?
I saw that you recommended changing the model in the thread so I changed it to O4 (Mini) but get the same issue.
And with the relevant document highlighted, I still got this:
Me: summarise this document
AI: Please upload the document or provide its text here, and I will gladly generate a summary for you.
When I say "highlighted’/selected, this is what I mean:
- Have you tried a different document?
- Summarization is available via Edit > Summarize via Chat. I would try that, especially since you’re not asking anything specific about the document.
PS: Why does your filename contain URLs, DOI, etc.??
I have tried it with lots of documents (I had this a while back and gave up, but thought I’d try again today - and had the same behaviour - I.e. random).
I do use prompts. to query specific aspects of the document (or a set of docs) - so the Edit-Summarise command, whilst useful, doesn’t allow you to engage with the document. Am honestly still puzzled - by its inconsistency as much as anything else.
The DOI was in the title because I had just copied citations details straight from the page I downloaded the document from. Most of my documents just have the title.
And in case it is of interest, here is my trying to ask the AI why I get this null response:
Me: why did you say you could not access the files and then you did access them. why are you inconsistent?
AI; I apologize for the inconsistency. Initially, I cannot directly access external links or files stored outside this chat environment. However, once you have selected or uploaded the document within this system, I can then analyze its content if provided or if I am given access to the actual text.
In this case, I was able to review the content because you indicated the documents were selected in your database, and I accessed the content through the system’s interface based on your instructions.
Thank you for pointing out the inconsistency—I’ll strive to clarify this better in the future. Would you like me to help further with this document or any other task?
Me: I want you always to operate on the documents that are highlighted in the database
AI: Understood. Moving forward, I will operate exclusively on the documents that are highlighted in your database. Please let me know what specific task you’d like me to perform on these highlighted documents, such as summarizing, extracting quotations, or analyzing content.
It then did a summary of a selection of documents. But it then failed to find documents with my next request (response was: Please upload the specific document or provide its text, so I can generate a summary for you.;
Maybe you are better of with DeepL or something like that. Not only because its translations are outstanding, but also because it will not overwrite anything afaict.
1 Like
Just tried this prompt successfully using GPT 4.1 (Nano) and a rich text and a PDF document. Only the web & Wikipedia search were enabled in Settings > AI > Chat
But especially local and cheap commercial models get easily confused if most or even all options are enabled. In this case a more precise prompt like Summarise the text of the selected document should usually work. In the end it’s all about the prompt…
Thanks - life is strange! So I changed the setting to what you set out (GPT 4.1 Nano) and only enabled the web and Wikipedia search, and used your exact prompt. Still nothing - it said the document was not found. Then I quite DEVONthink and reopened it…and it all works (and I can interrogate the document not just summarised). So I’m not sure what gremlin stops it working, but it is strangely fine now after a quit/start.
So problem solved for now! Will revert if I get any clearer on what causes the strange behaviour.
Thanks, I may try other providers than ChatGPT. I’ve found Google’s NoteBookLM to be really good as a browser AI - but it doesn’t allow an API unless you use the Enterprise version, which I don’t need.
As noted below, after quit/restarting DEVONthink the problem has gone (for now?). Strange, but to be expected I suppose in these early stages.
I’ve switched to Kagi Translate a while ago.
DeepL was something I was using for years, but I personally find the quality of the service has deteriorated recently.