Really enjoying the AI-Bridges Symposium (I'm attending virtually, from the Arcadia office in another part of London).

Exciting to be playing with the Wikidata MCP. Probably my first time actually consciously using MCP.

Curiously it hits the exact use case I have for LLMs -> generating better wikidata SPARQL queries.

Also first time playing with the Mistral LLM.

#AIBridges

'normal' mainstream LLMs appear to be unbelievably poor at writing Wikidata SPARQL queries. I'm hoping that connecting the WIkidata MCP will enable LLMs like Mistral to help write _better_ if not perfect SPARQL queries for me.

...and albeit this is an n=1 example lacking rigor but...

if you ask standard Mistral "Write a SPARQL query for Wikidata to find all planets that have more than two moons" it gives a very poor answer that doesn't give a sensible SPARQL query.

Whereas _after_ connecting the Wikidata MCP, Mistral with the very same prompt generates a Wikidata-informed SPARQL query which "works" (gives the expected/intended result in the Wikidata Query Service).

Interesting.