How close are we to AI generating ads for each individual viewer? A Wells Fargo commercial voiced over by Owen Wilson (one of the two impressions I can do) talking about a person named Oliver (that’s me!) trying to save money by eating before a game (something I am in the habit of) made me worried we were already in a world of per-person unique ads. We’re not, but I think we’re close.

Qwen released a new text to speech model a few days ago. The speech it generates isn’t natural sounding, but it’s realistic enough in short clips e.g. saying your name to get your attention before playing the rest of the ad. The model is small enough to run on device, so the voice generation could be offloaded onto personal devices to avoid data center costs (it’d be sad if on device ad generation is the real use case of all the AI-acceleration hardware on phones today).

Being targeted by advertising is not a good feeling, but we’ve all become used to it. We think nothing of being served ads for a skin care product we googled a week ago. I think it’s less likely that we’ll accept per-person targeting bleeding into the content of the ad. That’s a bit too creepy. But the past 30 years have shown that we trade privacy for convenience.