Full-bandwidth Transformer
A Transformer architecture that adds latent-space channels beyond the standard text output. The latent channels are not constrained by tokenization compression and have a far higher information capacity than text, allowing large amounts of parallel computation in latent space.
Timeline (last 90 days)
AI safety research organization Redwood Research published an analysis article pointing out that latent-space reasoning architectures represented by COCONUT and full-bandwidth Transformers may greatly weaken the currently most effective AI supervision tool — chain-of-thought (CoT).
The full-bandwidth Transformer retains standard text output while additionally adding a latent channel not constrained by the text bottleneck, enabling the model to perform large amounts of computation in parallel in latent space, with text presenting only a small portion of it.
All Facts (2)
AI safety research organization Redwood Research published an analysis article pointing out that latent-space reasoning architectures represented by COCONUT and full-bandwidth Transformers may greatly weaken the currently most effective AI supervision tool — chain-of-thought (CoT).
50%UnverifiedThe full-bandwidth Transformer retains standard text output while additionally adding a latent channel not constrained by the text bottleneck, enabling the model to perform large amounts of computation in parallel in latent space, with text presenting only a small portion of it.
50%