DeepTutor repairs the order linking document text to vectors
DeepTutor’s 1.7.3 release restores the original input order when an OpenAI-compatible service returns indexed embedding responses in a different sequence. The change addresses how document text is associated with its numerical representation. Previously affected indexes need rebuilding, while existing files are retained. The same release adds reviewable edits to generated books and changes to reading tools.
Artificial Intelligence··Midday
Returned vectors follow their original inputs
DeepTutor, a software project for AI-assisted learning and document work, corrects embedding handling in version 1.7.3. Embeddings are numerical representations associated with pieces of text. When a service compatible with OpenAI returns numbered responses out of sequence, DeepTutor places them back in input order. Indexes already affected by misplaced vectors require rebuilding; the update leaves the files themselves in place.[1]
Book changes can be reviewed before application
Generated books can be changed chapter by chapter or section by section without recreating the whole work. Proposed structural changes, including those made through page chat, display a comparison before they take effect. The last edit can be reversed while it remains unchanged by later work. The document reader adds continuous zoom, fitting text to the screen and an image viewer.[1]
Quiz answers stay on the server
Reading quizzes retain their answer keys on the server, while help connected to a question is stored separately from quiz attempts. Guardians with authorization can see progress, previous quizzes and help activity. For local helper agents, the initial Agent Client Protocol connection cannot handle interactive clarification or permission requests. If condensed conversation history still exceeds its planned bound, the chat stops with an explicit error.[1]