Click here for the second "About SolderSmoke" podcast.
http://www.soldersmoke.com/About SolderSmoke 2.mp3
This one looks not at the SolderSmoke Daily News blog, but instead at the SolderSmoke podcast itself.
I was delighted to see the inclusion of Pete, Dean, Farhan and Mike Rainey! This was really great.
Look, it is not perfect. There are errors. But probably about the same number of errors that you would get from real, human hosts, right? Voltaire told us not to let the perfect be the enemy of the good. And I find this to be amazingly good. Look, it is so good that it is kind of scary, right?
Before you get too critical realize how this has been done: I did nothing more than load the SolderSmoke Podcast Archive website into GPT-like model. Then I asked it to produce a deep dive podcast. That's it. About 5 clicks. It developed the podcast in about 3 minutes. I did the same thing yesterday but with the SolderSmoke blog. And this is only the beginning.
Posts
Woah.
ReplyDeleteN5ALO
Several studies have shown that if you feed a LLM (large language model) / AI tool its own output, it quickly produces garbage with zero information content. Since this is an experiment, try feeding it a transcript of the MP3 and see what happens - how many cycles before nothing is left of the original? I think we'll quickly conclude that humans are the best version of Soldersmoke.
ReplyDeletePretty cool! Missing the fun part of the human element. Definitely won't replace the good old team (Bill, Pete, and newest member Dean)
ReplyDeleteVery impressive AI. Just ~five clicks. Amazing. As to how it relates to Soldersmoke, this mp3 gives a newcomer to this site a quick overview in less than 12 minutes.
ReplyDeleteWhat AI does in medical applications is stunning!
“A hamfest, now there’s a term I haven’t heard for a while”. So many silly comments.
ReplyDeleteSo instead of replacing missing segments with AI (as noted in a certain April Fool's edition), are we now going to replace the entire podcast with AI? I might have thought it's another AF joke, except it's not April. Scary... Very scary. :-)
ReplyDeleteIf I didn’t know this was AI generated, I’d be fooled. However, I wouldn’t listen to it: too dry, no presence, and the “scripted” interactions and banter are “truly” artificial.
ReplyDeleteThis use of a LLM is instructive, and points to some of the dangers inherent in such technologies. The focus of the model is to produce content that pleases, because that’s how the process works. There is little regard for accuracy over inaccurate content: provided it meets the brief, it’s OK. It strives for “authenticity” but falls at the first hurdle IMO.
As I said as a comment to the previous post “don’t give up the day job”.