Google Releases Faster Text-Generation Model Using a New Approach

Google released DiffusionGemma in June, an experimental model that generates text differently than most AI systems. Instead of writing one word at a time, it refines entire sentences together, cutting response time significantly. This method is still experimental and faces hurdles with consistency and following instructions. The model is open for developers to test against conventional approaches in speed-critical applications.