LLM Title Fixes Summary | Generated by AI

Home 2026.05

Good. Here’s a summary of the three fixes applied:

Problem: The LLM ignored the “respond with only the title” instruction and returned 1518 chars of analysis instead of a 6-word title. The code then crashed with a ValueError.

Fixes (3 changes across 2 files):

  1. create_note_utils.py – max_tokens=60 on the API call: This is the key fix. Previously no max_tokens was set, so the model could ramble freely. 60 tokens is generous for 6 words but cuts off long explanations.

  2. create_note_utils.py – Retry logic (3 attempts): Instead of crashing on the first too-long title, it retries up to 3 times with a warning. Only raises ValueError after exhausting retries.

  3. create_note_from_clipboard.py – Stronger prompt: Changed from a long instruction to a concise prompt ending with \n\nTitle: – this primes the model to complete with just the title, which works better with many models than imperative instructions.

Back Donate