Opus 5, Voice AI, and Open Weight Models

**Beth Lyons** (0:03)
Hey, good morning. Good Monday morning, everybody. Great to see everyone in the chat. Hey, Jen. Hey, everyone else who's on their way.
It is Monday, July 27th, 2026 We are, I am Beth Lyons. This is Andy Halliday. We are your co-host for The Daily AI Show today. And it is episode 776, I believe. We'll be at 780 by the end of the week and 800 before our anniversary.

**Andy Halliday** (0:41)
Wow. What kind of invisible accomplishment will this add to my resume? I'm not sure.

**Beth Lyons** (0:51)
Well, one of the things, as I've told folks over the last couple of weeks, I have been working on getting full word at a time transcripts from all of our shows.
And I have that pretty well satisfied. But then what does that mean that we could do? If we did one clip for all 800 shows, we would double the number of videos that we have on our thing. If we did two clips or maybe like one clip for five weeks of the first five weeks, but four clips from one of the last ones. I mean, it just becomes its own very large corpus outside of what we think, which is this was the show as opposed to this was the section of the show that then can have its own legs in some ways.

**Andy Halliday** (1:50)
So yeah, I think it would be very interesting to use the semantic analysis of the transcripts to do a then and now kind of comparison. So here's what we said then.
And here's what we say now about that subject. And just to see the remarkable progression of development that we've experienced over the course of these three years.

**Beth Lyons** (2:19)
Yeah. Actually, you're making me think I'd need to go back and look at the kinds of questions because it would be interesting for us to answer from now, the questions that we were asking from them that like, hey, Andy, what do you think about blah, blah, blah? Right? And like, hey, Beth from 2024, thanks for the question. Here's what Andy from 2026 thinks. That kind of stuff. All right. Jude, Jeff, Jean, thanks so much for joining us in the chat. Before we started, Andy, we were talking about voice because Claude announced a voice interaction for Claude.
And I didn't get very far in testing it because what I want to be able to do in voice conversation is set up context. And I didn't try it in a project. I probably could do it in a Claude project where I have set up documents and a conversation that could seed it. But in OpenAI, I could this morning give an article. I have a skill called Newsify, which basically says, I'm going to talk about this on the show. So don't give me a big analysis. Give me some talking points that I can frame around that. Then I could go to Voice and have a conversation within that same thread. That was helpful for me and interesting. I was saying that I used to love and prep for the show that way. I'd do some prep and then I'd just have a conversation for a half an hour about things.
There's been a lot of reasons why I didn't come back to it.
But I may be coming back to that habit.
I am using the, I think Australian lady voice. I don't know which one that is. I can tell that when I'm talking and I stop, she'll go, hmm, to give a little time. And I could see like, all right, I'm either going to get used to the, hmm, or I'm going to say, yeah, don't do that. Yeah.

**Andy Halliday** (4:48)
So I have used Whisper Flow for probably over a year or so. And that's the, in fact, on my coding workstation, I don't even have a keyboard. So the only way I, well, on the laptop I do, but I don't have a separate keyboard like, you know, one of these, which is a nice way to type if you're going to type. So I don't really type anything. I sometimes have to go over to the keyboard on the laptop stand and make a change. So I use Whisper Flow. I've never been anything but irritated by the voice interactions I've tried on the phone or on the desktop. Largely because of the previous inability to really sense correctly when I'm finished with my exposition, if you will. If I pause, but it's a fragment of a sentence that I paused at, don't interrupt me.
I'm forming the compound addition to that sentence that I need to do. So the way I talk, as you can tell if you listen to me, it's very hard for AIs to understand in this smooth cadence that you expect in a back and forth conversation.