I wish he would have asked his questions to someone at Futuri and allowed then to answer them on the record for his article. They're doing promotions for RadioGPT right now, I'm sure they would have appreciated the opportunity to talk about it in a radio trade column.
-
Get involved.
We want your input!
Apply for Membership and join the conversations about everything related to broadcasting.
After we receive your registration, a moderator will review it. After your registration is approved, you will be permitted to post.
If you use a disposable or false email address, your registration will be rejected.
After your membership is approved, please take a minute to tell us a little bit about yourself.
https://www.radiodiscussions.com/forums/introduce-yourself.1088/
Thanks in advance and have fun!
RadioDiscussions Administrators
RadioGPT, AI hosts and the future of human on-air talent
- Thread starter Theater of My Mind
- Start date
I wish he would have asked his questions to someone at Futuri and allowed then to answer them on the record for his article. They're doing promotions for RadioGPT right now, I'm sure they would have appreciated the opportunity to talk about it in a radio trade column.
The final two paragraphs of Sean's piece:
"Radio’s AI discussion is exploding in time for convention season and, one hopes, in time for these questions to be discussed earnestly and as soon as possible. The purpose in asking them here is not to pose them of any one person, but to all of us. The discussion started between (Daniel) Anstandig and NAB’s Josh Miely on this week’s NAB podcast should continue in Las Vegas, and also at CRS in Nashville this month, at Radio Days Europe, and at the All Access Radio Summit.
Broadcasters, meanwhile, need to make sure they’re making the most of whatever resources they have. As we worry about chatbots taking on sentient personalities, we need to ask if our radio stations have those as well. That discussion will continue in this column."
It's all in the programming of the AI, though, isn't it? Certainly, the sources used could be limited to certain websites.
Then that wouldn't be considered true AI, but nothing more than scraping and reading selected text from limited websites or Blogs.
From one of the founders of AI starting back in the 50's, IBM: "At its simplest form, artificial intelligence is a field, which combines computer science and robust datasets, to enable problem-solving. It also encompasses sub-fields of machine learning and deep learning, which are frequently mentioned in conjunction with artificial intelligence. These disciplines are comprised of AI algorithms which seek to create expert systems which make predictions or classifications based on input data."
In the case of ChatGPT, which is by definition nothing more than a Predictive Speech Algorythm, same as other AI models being talked about in the news lately, successful true AI relies on a robust dataset (memories to humans) to solve problems, arrive at conclusions, or anticipate what to do next based on instructions. Currently, those memories are available solely by the public Internet, but just like Wikipedia on a wider scale, information on the Internet has greatly evolved from accurate, impartial news or data-sharing between colleges and government, to mainly financial or politically motivated information or opinions.
Perhaps to assuage @Kelly A a bit...
evertiq.com
"According to TrendForce’s estimation, the number of GPUs that the GPT model needed to process training data in 2020 came to around 20,000. Going forward, the number of GPUs that will be needed for the commercialization of the GPT model (or ChatGPT) is projected to reach above 30,000."
One would think that the implemented system is going to have text ready to broadcast (as audio) prior to when it's needed to go on-air. Whether that's seconds before, a minute before, or something else, nobody really has said.
EDIT: Insightful article (limited time availability for free...) https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html
GPU demand from ChatGPT could reach 30,000 chips
The recent arrival of ChatGPT has generated a lot of buzz across the industry sectors related to cloud computing and artificial intelligence (AI).
evertiq.com
"According to TrendForce’s estimation, the number of GPUs that the GPT model needed to process training data in 2020 came to around 20,000. Going forward, the number of GPUs that will be needed for the commercialization of the GPT model (or ChatGPT) is projected to reach above 30,000."
One would think that the implemented system is going to have text ready to broadcast (as audio) prior to when it's needed to go on-air. Whether that's seconds before, a minute before, or something else, nobody really has said.
EDIT: Insightful article (limited time availability for free...) https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html
Last edited:
I would think you could limit what information that AI uses (example might be trending topics from a local TV partner or a handful of trustworthy websites. It would be easy enough for it to search sports scores and updates (I even remember doing that with carts...."team loses play this, team wins play that").Perhaps to assuage @Kelly A a bit...
GPU demand from ChatGPT could reach 30,000 chips
The recent arrival of ChatGPT has generated a lot of buzz across the industry sectors related to cloud computing and artificial intelligence (AI).
"According to TrendForce’s estimation, the number of GPUs that the GPT model needed to process training data in 2020 came to around 20,000. Going forward, the number of GPUs that will be needed for the commercialization of the GPT model (or ChatGPT) is projected to reach above 30,000."
One would think that the implemented system is going to have text ready to broadcast (as audio) prior to when it's needed to go on-air. Whether that's seconds before, a minute before, or something else, nobody really has said.
EDIT: Insightful article (limited time availability for free...) https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html
From the article that @michael hagerty quoted:
RadioGPT is an extension of Futuri’s TopicPulse service which, as Cridland notes, is not untouched by human hands. The role of humans in helping create that content makes the use of AI voices not that different from the role of text-to-speech/speech-to-text that we’ve already accepted in our daily lives. As other groups and other vendors provide similar products, that will not always be the case and should be addressed now, including a protocol for human-sounding air talent names (or website bios) and live-sounding endorsements.
emphasis added
So, first generation will have guard-rails of some sort. Second generation might not. How fast that occurs will be the next story in the saga?
RadioGPT is an extension of Futuri’s TopicPulse service which, as Cridland notes, is not untouched by human hands. The role of humans in helping create that content makes the use of AI voices not that different from the role of text-to-speech/speech-to-text that we’ve already accepted in our daily lives. As other groups and other vendors provide similar products, that will not always be the case and should be addressed now, including a protocol for human-sounding air talent names (or website bios) and live-sounding endorsements.
emphasis added
So, first generation will have guard-rails of some sort. Second generation might not. How fast that occurs will be the next story in the saga?
I just got a telephone message from my church about the men's meeting.
Since the voice didn't sound familiar I suspected but now it's clear.
The voice said the meeting was at Bah-jangles. Everyone around here knows it's pronounced Boh. It's Bo Time!
That's a fast food chain in the South.
Since the voice didn't sound familiar I suspected but now it's clear.
The voice said the meeting was at Bah-jangles. Everyone around here knows it's pronounced Boh. It's Bo Time!
That's a fast food chain in the South.
I think there's a bit of a muddle on this thread - we've had the capability to do text-to-speech on computers for many years, I remember my 80s IBM PC being able to gabble in a very robotic computer voice, and for some years now a computer has been able to read a human-written script almost naturally.
What's changed in recent months is the ability for the computer to write the script in a variety of styles and registers. It can take various sources of information - music trivia, local events, weather, traffic - and write a presenter link, and then read that presenter link. It can react without human intervention to changes in circumstances - developing news from a news feed, a crash on a bridge from the Inrix API - and update what it's saying in real time.
Having said that, I've been unimpressed with the demos of Spotify's DJ that I've seen so far - it seems to be limited in what it says and keeps coming out with things like "so you've been listening to lots of 1990s dance music lately, here's some more 1990s dance". Hardly award-winning personality radio. I can see AI eventually developing to the point where it's replacing jockless hours on small-market stations, but radio's USP is the connection with the human personality, and AI can't do that just yet.
All great points. For some reason, folks here, and I'm sure as seen on other media like video gaming, insist on merging capabilities of synthesized speech, with some form of AI to create models like the DJ's of their past. It just doesn't work that way.
Not only that, but radio is one of the oldest forms of media that's adapted to evolving public media consumption, not by doing more of the old way, but working to give the wider audience what they want within the bounds of a one-way transmission at the lowest possible cost. The other consideration that commenter's convinced that AI will bring back radio personalities to fool the public don't understand; is that radio as an industry is inherently cheap, especially as it comes to adopting cutting edge technology. Common voice tracking today, is the most cost effective way to break up long sets of music, get in or out of breaks, or do station branding. It's common to have someone at a station within a group doing weekend or evening voicetracks for a dozen or more other stations around the country every week. Having some synthesized speech API that's scraping content from the public Internet, would be much more expensive to do, sound obviously fake, and would likely provide no financial advantage.
I think that's correct. If for no other reason, it would require some serious vetting. Just let an AI DJ riff live on the air and who knows what could happen!
Then the lawyers would have to figure out who to sue.Here's an opinion piece calling for regulation of AI:
thehill.com
How dangerous is AI? Regulate it before it’s too late
This technology feels like a runaway train that we’re chasing on foot.
thehill.com
From the demo it appears that what the AI says is heavily regulated, and you'd do the same thing at a station level, just like with any other talent. "Liner #3 into trending topics"I think that's correct. If for no other reason, it would require some serious vetting. Just let an AI DJ riff live on the air and who knows what could happen!
But how do you sue a robot? But I guess if the robot uses AI algorithms to invest in the stock market, he/she/it probably has deep pockets!
You don't sue a robot. You sue the company that's making money from licensing the robot.
But suppose the company that makes the robot is run by a robot?
Of course I'm getting ahead of myself and I guess my sense of humor isn't as obvious as I thought.
- Status