Fix privacy policy for AI summarization

InAnYan / jabref

Graphical Java application for managing BibTeX and biblatex (.bib) databases

https://devdocs.jabref.org

MIT License

0 stars 0 forks source link

Fix privacy policy for AI summarization #92

Closed koppor closed 3 months ago

koppor commented 3 months ago

[x] (1) shows "AI summary", but (2) shows "chatting". Is the Privacy Policy right?
[x] (3) shows missing dots. Please add a dot after each sentence.
[x] If possible, there should a link to "Deep Java Library" in the privacy policy
[x] Add information how much will be downloaded (300 MB?)

InAnYan commented 3 months ago

DJL said that the downloads are anonymous and no data is collected. Is that okay to mention in Privacy Policy?

InAnYan commented 3 months ago

For downloading size: it depends on selected embedding model and, I'm not sure of a way to estimate download size... Different models weight differently

koppor commented 3 months ago

DJL said that the downloads are anonymous and no data is collected.

Via email?

Can you ask them if it was OK to make that email public?

Is that okay to mention in Privacy Policy?

We need to have a "trusted" source. If a company sues JabRef e.V. we need to make sure that we did our best to ensure that the statements we do are true. If we just say: "It was claimed privately to some of our contributors", this is not enough.

It is still bad that they don't have a privacy policy, but we need to accept.

The alternative is that we mirror the models...

koppor commented 3 months ago

For downloading size: it depends on selected embedding model and, I'm not sure of a way to estimate download size... Different models weight differently

Just estimate 1 GB download size for each.

Alterntive:

HTTP allows to get the file size before downloading: https://stackoverflow.com/questions/122853/how-to-get-the-file-size-from-http-headers -- Maybe, the servers of them report the size back
hard-coded store the current known size inside jabref. At runtime, look it up. 😅

InAnYan commented 3 months ago

https://github.com/deepjavalibrary/djl/discussions/3370#discussioncomment-10233632

What should we do?

koppor commented 3 months ago

deepjavalibrary/djl#3370 (comment)

What should we do?

We can link these statements.

InAnYan commented 3 months ago

Nearly got mad.

Neither it's understandable how they download models from the code,

neither from Wireshark....

I guess hardcoded values would be better. Embedding model is not something that is actively updated. If it's updated, that would be a new model

koppor commented 3 months ago

Nearly got mad.

It will be especially interesting if one aims to package all known components to Ubuntu (without using Flatpack or Snap) or NixOS. Then, all files must be available on other servers ^^.

InAnYan commented 3 months ago

Phew, what do you think of recent push?

koppor commented 3 months ago

It is OK. However, consistency is an issue. I think, I did not tell often enough that consistency in a UI is important.

[x] Change the sentence to: "You find information about the privacy policy here".

Then, it is somehow similar to the sentences above, even if all text is blue. This is OK. On the first look, the new proposed sentence is similar to the other ones.

[x] Add a space after the sentence before. Currently, there is no space between the dot and the link.

koppor commented 3 months ago

Moved to week 1, because the "Privacy notice" also appears at "AI chat"

InAnYan commented 3 months ago

Change the sentence to: "You find information about the privacy policy here".

I thought about that, was a little worried, that user will expect link after word "here"

InAnYan commented 3 months ago

I think I'll close the issue, as I fixed everything you wrote