Overview
- In early May the Hangzhou startup Meng Xiaoyi opened preorders for a 27‑gram collar called PettiChat and reported about 10,000 reservations ahead of a planned end‑of‑May sale.
- The company says the device uses built‑in microphones, motion sensors and Alibaba Cloud’s Qwen model to turn barks, meows and posture into human‑language phrases in about 1.2 seconds.
- Meng Xiaoyi advertises near‑95% accuracy and claims the model was trained on millions of pet audio samples and millions of labelled data points to identify more than 20 emotions and behaviours.
- Journalists, veterinarians and engineers point out the company has not published datasets or peer‑reviewed tests and critics say the promotional videos look staged and could have post‑production edits.
- If the product reaches consumers as marketed it could reshape pet tech demand but it also raises risk of misinterpretation for owners and underscores the need for third‑party validation, transparent methods and clearer welfare safeguards.