Perplexity AI Initiates Research to Develop Automated Scientist with 10,000 GPUs
Perplexity, an AI search engine, has launched a new Frontier Research initiative aimed at building an automated scientist. This ambitious project commits at least 10,000 GPUs to focus on multi-agent orchestration and continual learning systems. The company's goal is to enable AI to produce scientific understanding and new knowledge, moving beyond mere information retrieval or prediction. Perplexity defines research as a cycle involving choosing unanswered questions, examining existing knowledge, identifying gaps, forming hypotheses, running experiments, and revising explanations. The proposed system will utilize various AI agents, some searching thousands of sources, others investigating specific threads, suggesting experiments, or critiquing hypotheses. The initiative will explore whether generated training data can outperform real data and how small training experiments scale to larger runs. Perplexity plans to publish evaluations, papers, blogs, and open-source its models and tooling.