#Voice recognition changed how I think

One part of AI that is deeply changing the way I think is voice recognition.

As everyone has noticed, speech understanding has taken a huge leap since LLMs arrived, making the old systems look dated. Much of that comes from the models being multimodal: they no longer just turn sound into words. They grasp the overall context of what you’re dictating (the subject, your intent, the likely phrasing), which lets them clear up ambiguities and give you coherent text even from poor audio.

If you’re someone who always has ideas running through their head, it’s life-changing. You used to pull out your phone and type painfully on a tiny keyboard, or scribble in a Moleskine. Then, once you were sitting at a desk, go back to your notes, dig in, look things up. Now the gap between an idea popping up and an answer has shrunk dramatically: you just talk to ChatGPT. It gets nearly everything and answers on the spot. Writing becomes optional.

For me, it means I follow up on far more hunches. I no longer have to jot them down in a rush before they evaporate.

Still, don’t try chatting with ChatGPT on a Vélib (Paris’s shared bikes). Last time I ended up on the ground, and the cobblestones have no conversational interface.