bidirectional

This paper presents a mechanistic analysis of induction in masked diffusion language models, identifying a bidirectional induction circuit and showing that these models use the global fraction of masked tokens as an implicit timestep.

Flex-Forcing introduces a unified framework for video diffusion that supports both autoregressive and bidirectional generation modes, offering flexible control for video generation tasks.

Introduces Flex-Forcing, a unified training and inference framework that allows video diffusion models to operate under both bidirectional and autoregressive regimes via a flexible chunking mechanism over temporal and denoising steps, achieving better video quality, long-video stability, and faster inference.

OpenAI's new voice model Bidi 1 first test exposure, supports bidirectional voice design, real-time translation, and stronger context memory, currently being pushed to a small group on ChatGPT.

OpenAI is rolling out a new bidirectional voice model (Bidi 1) for ChatGPT that allows simultaneous speaking, hearing, and listening, real-time translation, and improved conversation context handling. The upgrade is appearing in the web interface and app for some users, with a broader release expected soon.

A user asks which upcoming OpenAI feature is more exciting: the rumored GPT-5.6 model or bidirectional voice mode (BiDi), which allows real-time simultaneous listening and speaking.

An example of the upcoming GPT bidirectional voice model has been shown.

OpenAI plans to release GPT-Bidi-1, its next-generation voice model that can listen and speak simultaneously, handle interruptions, and enable more natural conversations.

Introduces BES (Bidirectional Evolutionary Search), a search framework for LLMs that combines forward candidate evolution with backward goal decomposition to improve sampling on hard reasoning problems during post-training and inference.