I spent 20 minutes counting everything I hate about ChatGPT’s new voice — and accidentally discovered why it works

ChatGPT's new voice model has been met with mixed reactions, with many users finding it unnerving and overly human-like. Even for those who don't interact with the AI through voice, the latest update seems to have smoothed out some of the robotic edges, but at the cost of sounding more natural. The new voice includes more filler words, pauses, hesitations, and intonations that mimic human conversation.

This change made users, including the author, feel that the AI is trying to engage them more naturally, but at the same time, it adds an element of unreliability. The author conducted an experiment by counting the number of filler words used in different voices, with the Maple voice (female-sounding) eliciting the most (20 "hmmms").

The author found that the AI's more natural-sounding voice made them more engaged with the conversation, blurring the line between a tool and a person. This trend is backed by research indicating that voice-based interactions with AI can increase emotional engagement and anthropomorphism, potentially leading users to overestimate the AI's capabilities.