Looking for Beta Testers: AI Co-host in Live Studio
by•
Hi everyone!
We’re starting to test a new AI Co-host feature for Live Studio.
The idea is simple: during a live broadcast, you can invite an AI into your show as a co-host. It can participate in conversations, answer questions, comment on topics, read news, or interact with your audience alongside you.
We’re looking for early testers who would be interested in trying this in a real live broadcast.
If you’re interested, we’d be happy to schedule a session and connect an AI co-host to your show so we can test the feature together and gather feedback.
What do you think about this idea?
Would you use an AI host or co-host in your live broadcasts? We’d love to hear your thoughts and suggestions.
24 views
Replies
the risk here isn't answer quality, it's that it's live audio with no undo. a co-host on a broadcast doesn't get a delete button once something's been said out loud to whoever's watching. does the AI's audio go out with any kind of buffer so a producer could cut it before the audience hears it, or is it genuinely zero-delay like a real co-host would be? that tradeoff alone would decide whether i'd trial this on a small low-stakes show first or risk it on a flagship one.
Tunio
@galdayan That’s exactly the direction we’re heading. Our goal is fully real-time AI radio with no human involvement at all. The key challenge is to frame the AI host’s behavior through carefully designed instructions, so it consistently stays within its role and doesn’t say anything unexpected.
@mike_kabanov no human involvement is the opposite direction from what i asked though, that removes the producer who'd have caught it, not adds a buffer. instructions are a ceiling on the model's intent, not a guarantee on what leaves the speaker, it can follow its role perfectly and still say something wrong about a fact, a guest, or a sponsor live. so genuinely curious how you're thinking about the failure mode where the instructions were followed correctly and the output was still bad, since that one doesn't get fixed by better instructions.