Long live Pied Piper.
Luka, a messaging app that builds all kinds of chatbots that do things like make restaurant reservations, rolled out three new bots this weekend, each mimicking a different character from HBO’s Silicon Valley.
That means you can finally get advice from top-notch investors and entrepreneurs like Russ “Three Commas” Hanneman or Erlich Bachman, the coattail-riding “co-founder” of the show’s main startup, Pied Piper.
Luka built the bots by analyzing all of the characters’ lines from the first two seasons of the show, said Luka CEO Eugenia Kuyda. The language is then dumped into a “neural network dialogue model” that learns the characters’ language patterns.
In other words, the bots are meant to feel like real people. They aren’t limited to a dozen canned responses, but can actually say things the character might say in any given situation, even if they’ve never said that sentence or phrase before. At least that’s the idea.
It sorta works, too. When I asked Bachman for a pizza recommendation this weekend, he replied, “I don’t give a f*ck, man. Go ask Yelp.” Sounds about right!
The bots are a well-timed gimmick — Silicon Valley’s season 3 premiere airs on HBO tonight — but the idea and the tech behind them are worth taking more seriously than Bachman’s trash talk. That’s because Luka (and presumably other bot makers) want to make bots that mimic real-life people, not just fictitious television characters.
Imagine a bot that can pick up your conversational habits and personality and then talk to someone else on your behalf. Luka built these Silicon Valley bots on a relatively small amount of conversational data. In Hanneman’s, case it was just over 1,000 lines of text.
“Think about those who tweet, speak and write every day,” Kuyda wrote on Luka’s blog. “Their AIs will be significantly more powerful.”
So watch out. Your bot is coming to replace you.
If you want to chat with the Silicon Valley characters, you can download Luka from the App Store for free and test them out for yourself.
This article originally appeared on Recode.net.