Closed Bug 1903900 Opened 2 years ago Closed 1 year ago

Better handle different chat provider GET request limits for long text selection

Categories

(Core :: Machine Learning: General, enhancement, P1)

enhancement
Points:
2

Tracking

()

VERIFIED FIXED
133 Branch
Tracking Status
firefox132 --- wontfix
firefox133 --- verified

People

(Reporter: Mardak, Assigned: Mardak)

References

Details

(Whiteboard: [genai])

Attachments

(1 file)

With bug 1895433, currently just documenting in comments approximately how large of requests each provider accepts with a fixed text selection truncation. We could be smarter by allowing longer or shorter text selection depending on which provider is active and/or at least better inform user it'll be truncated.

See Also: → 1905898
Blocks: 1900914
No longer blocks: 1894996
Blocks: 1911129
No longer blocks: 1900914
Points: --- → 5
Priority: -- → P1
See Also: → 1916621

Still approximate but slightly smarter per-provider prompt max length limits.

Assignee: nobody → edilee
Status: NEW → ASSIGNED
Points: 5 → 2

[Tracking Requested - why for this release]: avoid poor ux of chatbot seemingly load nothing on long selection for some providers

Pushed by elee@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/408add1d248d Better handle different chat provider GET request limits for long text selection r=tarek
Flags: qe-verify+
Status: ASSIGNED → RESOLVED
Closed: 1 year ago
Resolution: --- → FIXED
Target Milestone: --- → 133 Branch

Hi @Ed, How can we verify this fix ? what is being implemented here ? Should there be a limit of how much text are we selecting ? or should we not see the shortcut if too much text is being selected ? with certain providers If We select a lot of text we still get the Server error from this issue Bug 1905898

Flags: needinfo?(edilee)

It should be selecting text and choosing a prompt just like bug 1905898. I tested with https://en.wikipedia.org/wiki/Firefox selecting from the beginning then making sure there's no server error in the sidebar as well as the text sent to the provider is more (for gemini) or less (for lechat) by looking for the wikipedia [12] vs [34] citations.

For reference using that wikipedia page, mouse down, cmd-a select all, mouse up to get the shortcuts icon and selecting summarize:

  • claude includes [78] ending with permission. This happened
  • chatgpt includes [78] ending with permission. This happened
  • gemini includes [256] ending with Support Release (ESR)
  • huggingchat includes [41] ending with "rapid" one; by
  • lechat includes [18] ending with "Phoenix" by members

I did test on some other wikipedia pages and noticed it sometimes giving server errors (on chatgpt and lechat), so maybe we'll verify this bug as different lengths per provider, e.g., those wikipedia 18 vs 78 vs 41 vs 256 cite numbers, and we'll use bug 1905898 to better avoid the server errors.

Flags: needinfo?(edilee) → needinfo?(rdoghi)

pushing experiment to 133

Thanks Ed. This is Verified as fixed in our latest Nightly 133.0a1 (2024-10-02) the Claude Ai did go to [79][80].

Flags: qe-verify+
Flags: needinfo?(rdoghi)
See Also: → 1922379

Updating the main status flag.

Status: RESOLVED → VERIFIED
See Also: → 1916450
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: