"We Are Very Early in Our Work With LLMs," - Prem Ramaswami, Head of Data Commons at Google

Google Data Commons launched an MCP server to ground AI in verifiable public data from trusted sources like the UN, World Bank, and Census Bureau. The clever part: users' own LLMs do the translation work, so Google's compute isn't involved. Prem Ramaswami argues we're still "very early" with LLMs (Google's transformer paper was only 2017) and the answer to hallucinations is "try all of the above" - combining language models with robust, auditable data sources. The service is free, integrates hundreds of datasets with transparent provenance, and chose Anthropic's open MCP standard over building proprietary infrastructure. Key challenge: expanding beyond strong US/OECD coverage to make grounded AI systems globally representative.Retry