Ask HN: Why Is Siri AI So Poor at Basic Tasks?

Summary

A Hacker News user with experience using large language models questions why the new Siri performs poorly on basic cross-app requests. In one example, Siri surfaced a five-day-old message about a delivery driver instead of a recently received text containing the driver’s location. In another, it created an appointment for the wrong day and time, then moved a different appointment after the user corrected it. Siri also asked what a message should say after being told to text Jane that the user was on the way home. The user says older Siri handled some of these tasks correctly and argues that a basic LLM with tool access should manage them. The post questions whether LLM technology is being used effectively and asks why performance remains weak if Gemini is the backend.