## Lecture Video
### New Features
* Use the chat interface to ask an assistant questions about the Lecture
Video being played.
* Chat in Lecture Video mode requires a Version 2 manifest that includes
`word_level_transcription`.
* The Lecture Video will pause as soon as users start typing in the chat
interface.
* Sending messages is only allowed when a Lecture Video is playing, or
after having answered a question.
* PingPong will send representative samples of the Lecture Video
transcript, along with representative video frames, to provide the
necessary context to the model.
* After the Lecture Video has concluded, read a transcript of the entire
conversation.
### Known Issues
* Selecting an assistant model is not currently supported. PingPong will
always use the assistant default model, which is currently GPT-5.4.
* Specifying instructions, reasoning effort, temperature, and other Chat
mode settings is not currently supported.

## Messages
### New Features
* Support for hidden messages that are filtered out from `get_thread`
and similar endpoints, but passed on to the model.

## Threads
### Updates & Improvements
* Updated auto-scrolling behavior, where the thread will only "snap" to
the bottom of the thread when near the bottom of the conversation only
when streaming a response.