OpenAI researcher: LLMs lack true omni output and have weird knowledge holes

willdepue · x · 2026-09-22

Continuing his list to Nat Friedman, OpenAI's willdepue notes LLMs lack true omni output (GPT voice can't make sound effects or beatbox; music gen uses non-frontier models) and have odd knowledge holes — they're surprisingly bad at recommending movies/music, having often not watched or read them.

Related event: OpenAI Researcher Lists LLM Shortcomings: Real-Time Voice, Context, Full Multimodality(3 posts)→

Original post →

More from Models

Models channel →