Google Meena: the 2.6B end-to-end chatbot

In one sentence Google introduces Meena, a 2.6B-parameter conversational model trained on 341 GB of social dialogue, along with SSA, a new metric for evaluating chatbot quality.

Verified Official source

ShareLinkedIn X

Before ChatGPT, most chatbots were rule-based or full of duct tape: they handled a handful of phrases well and everything else badly. Google tries a different approach: take a very large model and train it only to continue real conversations harvested from social media.

The result is Meena. It can talk about anything, keeps the thread over many turns, and produces replies that often read like a real person wrote them.

It's not an assistant yet: no search, no tools. But it's the first serious sign that the future of chatbots isn't rules, it's scale.