Apologies for doing another one of these so soon after the other one, but when it rains it pours.
- Since I last posted OpenAI released “advanced voice mode” to all plus subscribers. What this means is that the lag we’re used to when talking to AI is now gone, and now it is indiscernible from speaking to a real human being, up to and including detecting sarcasm, humor, and the like. I have been using it to brush up on my very rusty mission Spanish, and now any pre-missionary who wants to go above and beyond and practice, say, giving a first discussion in French with a personalized tutor that will correct their grammar doesn’t have to wait until they enter the MTC. They have put some safeguards in place so that it can’t just clone your voice, but the day when anybody can clone anybody’s voice and automate a thousand bots to call everybody in your phone is coming, so once again please be aware and discuss with your old and sometimes not so old-relatives that a phone call from somebody that sounds just like you asking for money isn’t necessarily you.
- Google is still very much playing catch up in the AI wars (and no matter how good they get, their AI will probably always invoke images of Black Nazis). Notebook LM has a fun new functionality that automatically generates a podcast-type back and forth based off of raw content that you feed into it. There’s even a pretty-decent Spotify podcast that is just somebody loading in different articles from the Stanford Encyclopedia of Philosophy. I created a podcast on the Bear Lake Monster below by feeding in all the Mormonr primary sources on the Bear Lake Monster as well as the Wikipedia article. I could detect some slight hallucination (sometimes it went into slightly more detail than was available in the content I fed into it as far as I could tell). Also, it seemed to draw mostly from the Wikipedia article, probably because the Mormonr PDFs weren’t as neatly formatted for AI digestion. But still, it produced a podcast that’s probably in the 50th or so percentile in terms of podcast production quality, and it did it in about five minutes.
Also, for fun I fed in the John Gee BYU Studies article on the Great Apostasy that was the subject of some discussion in my last post. There was a lot more content, so the podcast is quite a big longer, but again all that I did was drop the PDF in and wait for ten minutes. (Obviously it takes its cue from the article in terms of perspective, so if you don’t like the article you won’t like the podcast). I should note that the podcasts generated this way from the Stanford Encyclopedia were better (for those I’d put the podcasts at the 70th percentile), but that might be due to the subject matter playing better with AI production.
Podcast on John Gee’s article about the Great Apostasy
Podcast on the Bear Lake Monster
Wow. I think I’m in love with one of those fake podcasters. ;-D
Thanks for sharing. I just mapped out a podcast series that uses the same tool. I’m interested to see whether people will like it.
@ Ken: We laugh, but I think that is increasingly going to be an issue. At least it’s not a Scarlett Johansson voice like the first Chat-GPT advanced audio.
@ Sterling: It definitely works better for some subjects and content than others.