DeepMind researchers just introduced OPRO, a technique that turns large language models into natural language optimizers. Instead of writing complex math formulas, users provide a meta-prompt with task descriptions and prior scores, guiding the model to iteratively discover better prompts and boost task accuracy.
Google CEO Sundar Pichai just revealed that Bard will soon be upgraded to Google's larger PaLM models, bringing sharper reasoning, coding, and math capabilities. Addressing criticism that the bot lags behind rivals, Pichai admitted the initial launch on a lightweight LaMDA was like putting a "souped-up Civic" into a supercar race, but updates are rolling out next week.
Researchers just introduced PaLM-E, an embodied multimodal language model that injects real-world sensor data directly into an LLM's embedding space. Scaling up to 562B parameters, it turns high-level prompts like "bring me the rice chips" into multi-step physical manipulation plans, even generalizing to objects the robot has never encountered before.
