OpenAI researcher Will Depue on why voice models still lack true realtime chat

willdepue · x · 2026-09-22

OpenAI's Will Depue lists gaps in today's chat paradigm: no true bidirectional interruption despite omni voice models, 'chats' instead of one continuous stream, uneditable messages, GPT Voice unable to make sound effects or beatbox, and odd knowledge holes like models being worse at recommending movies/music than expected.

Related event: OpenAI researcher lists LLM shortcomings: context pollution, missing multimodality(3 posts)→

Original post →

More from Models

Models channel →