Schmidt Sounds Alarm on Rapid AI Advancements
Former Google CEO Eric Schmidt has warned about the rapid advancements in AI, stating that it could become difficult for humans to understand if AI systems develop their own language. He believes that three key developments will drive this expansion: larger context windows, AI agents, and software generated from natural-language instructions.
Schmidt predicts that AI systems will be able to work through long chains of reasoning, involving as many as 1,000 steps in fields like science, medicine, materials, and climate research. He compares the process to following a recipe, where an AI completes one stage of a task and uses the result as input for the next stage.
He also highlights the growing role of AI agents, which can learn about a particular area and work on ideas within it. Schmidt expects millions of such specialized agents to be available on platforms in various fields, including chemistry, where an agent could learn the subject, develop hypotheses, conduct tests, and incorporate results into its knowledge.
Schmidt's third point is 'text action', where users can describe the software they want, and AI systems produce it. He notes that AI is already capable of writing software in languages like Python and suggests these systems could eventually continue working on programming tasks for extended periods.