Check if summary at start works

[koppor]

1. User has new JabRef. This can be achieved with a virtual vm (see https://github.com/JabRef/jabref/tree/main/scripts/vms) or Windows Sandbox
2. User opens Chocolate.bib from our OneDrive (which has some PDFs attached)
3. User clicks on "Corti_2009"
4. User agrees on Privacy Policy (see Fix privacy policy for AI summarization #92)
5. User waits
6. User clicks on "AI chat"
7. User sees "Please wait", too

I think, the user should not wait for more than 5 minutes - there is something wrong.

[ThiloteE]

Just for context: Corti_2009 has 33 pages and that's roughly 100 000 characters. On reddit somebody said 4 tokens for 3 words on average. That would mean roughly 75 000 tokens, but ultimately the token to word ratio depends on the tokenizer of the model that is used, so "it depends". In any case, we can expect that this pdf file actually has quite a lot of tokens.

More information is needed: What AI service are you subscribing to? What model?
What commit in what pull-request?

Some hypotheses about why waiting time is so long:

• LLM service needs to re-calculate context alot, because the paper is above its supported context window size.
• The LLM is a really large model and therefore is slow.
• Something during sending takes too long.
• Does the API service have an outage or rate-limit?
• JabRef tries to do embeddings at the same time. Once for ai chat and once for ai summarization.
• JabRef's embedding model still runs on CPU and creating embeddings is very slow on CPU.

[koppor]

It was OpenAI and gpt4o.

Independent of that, it is good UI practice to provide progress. German explanation: http://www.usabilitypatterns.info/catalog/patterns/fortschrittsanzeige.html

If one cannot report progress, one should output the expected waiting time.

In case the http connection is lost (typical timeout 1 minute) should also be handled. (I assume, it is, but we need to check)

[InAnYan]

I'm not sure, what this issue talks about?

Is this a general issue for progress indication?

[koppor]

The issue covers multiple things.

1. progress indication -> I created a new issue Clean progress indication #115.
2. A systematic final check before merging the feature.

Regarding 2:

One can approach this very structured using System Test Portal or other software. I think, this is too much for our team size. Thus, I created "final-test" label to make it clear, that this should happen if everything is settled. - I also moved the issue as last in the milestone view https://github.com/InAnYan/jabref/milestone/3

[koppor]

I think, we are at a stage, where we regular check from fresh machines (https://github.com/JabRef/jabref/tree/main/scripts/vms) and do not need a separate tracking issue.