August 11, 2025 • 15 mins
OpenAI Plans $500 Billion Stock Sale to Enable Employee Cash Out Before IPO
OpenAI Unveils Affordable, Customizable Language Models gpt-oss-120b and gpt-oss-20b
Claude Opus 4.1
Google's Gemini 2.5 Deep Think Model Now Available for AI Ultra Subscribers
ElevenLabs Launches AI Model for Commercially Cleared Music with Industry Partnerships
Why AI Podcasts Are Failing to Win Audiences
#AI, #ChatGPT5, #OpenAI, #LanguageModels, #ClaudeOpus, #GoogleGemini, #AIPodcasts
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Welcome to Innovation Pulse, your quick no-nonsense update on the latest in AI.
(00:10):
First, we will cover the latest news.
Open AI is eyeing a $500 billion valuation with stock sales and new model releases,
while Claude Opus, 4.1, and Google's Gemini 2.5 are pushing coding boundaries.
After this, we will dive deep into the fascinating evolution of AI in podcasting,
(00:32):
examining how technology enhances creativity without replacing the human touch.
Open AI, the company behind ChatGPT, is considering a stock sale that could value it at around $500 billion,
up from its current valuation of $300 billion.
This sale would allow employees to cash out, highlighting Open AI's rapid growth in users and revenue,
(00:57):
driven by its flagship ChatGPT product.
Open AI's revenue has doubled in the first seven months of the year, with projections to hit $20 billion by year-end.
The company has seen a rise to 700 million weekly active users.
The stock sale discussions follow a funding round led by Softbank, aiming to raise $40 billion.
(01:22):
Tech companies are fiercely competing for AI talent, with meta-investing heavily in scale AI.
Open AI may shift from its capped profit model, potentially setting the stage for an IPO in the future.
Open AI has unveiled two new open-weight language models,
(01:43):
GPT-OSS, 120B and GPT-OSS, 20B, marking the first release of such models since GPT-2.
In 2019, these models are cost-effective, customizable, and designed for developers and researchers.
Unlike open-source models, open-weight models share their parameters publicly,
(02:06):
offering transparency but not complete source code access.
The release, delayed for additional safety testing, includes measures to filter harmful data and prevent malicious fine-tuning.
Open AI collaborated with companies like NVIDIA and AMD to ensure compatibility across various hardware.
(02:27):
The models available on platforms like Huggingface and GitHub can be used on consumer devices or cloud services.
Open AI aims to democratize AI access, allowing users to employ the models for tasks like advanced reasoning and personal assistance.
(02:48):
Join us as we step into the next evolution of coding excellence.
Claude Opus 4.1 is now available, enhancing real-world coding, reasoning, and agentic tasks.
It's accessible to paid Claude users and via Claude code, with no change in pricing from Opus 4.
The upgrade is available through the API, Amazon Bedrock, and Google Cloud's Vertex AI.
(03:14):
Opus 4.1 achieves a 74.5% coding performance on the SWE bench verified and improves research and data analysis skills.
Focusing on detail tracking and agentic search, GitHub highlights notable advancements in multi-file code refactoring, while Raccoon Group praises its precision in large code-based corrections.
(03:39):
WinSurf reports a significant improvement, comparable to past model upgrades.
Developers are encouraged to upgrade to Opus 4.1, accessible via Claude Opus 4.1, 2020 and 250805.
Feedback is welcome to aid in developing future models.
(04:02):
Google has introduced its most advanced AI model, Gemini 2.5 DeepThink, but access is limited to those subscribing to the $250 AI Ultra plan.
Released in the Gemini app, DeepThink is designed for complex queries, utilizing more computing resources than other models.
It enhances the foundation of Gemini 2.5 Pro by increasing thinking time and employing greater parallel analysis.
(04:31):
This allows it to explore multiple problem-solving approaches, revisiting and remixing hypotheses for higher quality outputs.
Although response times are longer, DeepThink excels in design aesthetics, scientific reasoning and coding.
It has undergone standard benchmarks, outperforming Gemini 2.5 Pro, and rivals like OpenAI 03 and Grock 4.
(04:57):
In a challenge called Humanity's Last Exam involving 2,500 complex questions across 100 subjects, DeepThink achieved a score of 34.8%, significantly surpassing other models.
11 Labs has unveiled a new AI model that generates music cleared for commercial use, expanding its focus beyond AI audio tools.
(05:24):
Known for its text-to-speech products, 11 Labs is now venturing into music, offering samples like a synthetic voice, rapping in a style reminiscent of artists like Dr. Dre and Kendrick Lamar.
This raises concerns about the source material for AI music training, as seen in lawsuits against Suno and Udio by the Recording Industry Association of America for using copyrighted music.
(05:50):
11 Labs has partnered with Merlin Network and Cobalt Music Group to license music for AI training. Cobalt ensures artists must opt in, offering benefits like new revenue streams, revenue sharing and safeguards against misuse, according to a Cobalt representative.
And now, pivot our discussion towards the main AI topic.
(06:20):
Hey everyone, Alex here with Innovation Pulse, and today I've got Yakov Lasker with me, who just spent way too much time diving into one of the strangest contradictions in tech right now.
Yakov, before we even say what we're talking about, hit me with that number that made my brain hurt.
Okay, so picture this. The AI podcasting market is exploding at 28% growth per year, heading toward $26 billion by 2033.
(06:48):
Meanwhile, when researchers actually tested human versus AI-generated podcasts, 96% of people chose the human version. 96%, Alex.
Wait, hold up. So we're throwing billions at AI podcast technology that people actively don't want?
Exactly, and here's the kicker. About 45 million Americans listen to AI-driven podcasts every month. But when you dig deeper, they're not actually listening to robot hosts. They're listening to humans who use AI tools for editing and production.
(07:18):
Oh, that's like saying people love AI music because they stream songs that were auto-tuned. The AI is behind the scenes, not replacing the artist.
Bingo, and this disconnect is teaching us something huge about how humans really interact with technology, especially when it comes to storytelling and connection.
Okay, so before we unpack why people are rejecting AI hosts, let's set the stage here. This isn't just about podcasting, right? This is happening while AI is taking over everything from writing to art to customer service.
(07:50):
Right, and podcasting seemed like the perfect candidate for AI takeover. I mean, it's just talking, right? We've got voice synthesis that can clone your voice from 10 seconds of audio.
Google's Notebook LM can turn any document into a conversation between two hosts that sounds incredibly realistic.
I've heard those Notebook LM podcasts. They're honestly creepy how good they sound, but you're telling me people still aren't buying it?
(08:14):
Here's where it gets fascinating. When researchers dug into the psychology of podcast listening, they found that podcasting succeeds because it creates what they call parasocial relationships.
Listeners feel like they actually know the hosts, like they're friends hanging out.
Oh, that makes total sense. I've definitely caught myself thinking, I wonder what Joe Rogan would say about this during random moments in my day.
(08:39):
Exactly. And here's the thing, AI can't authentically participate in that relationship because it has no genuine experiences to share.
It can't tell you about the time it bombed a presentation or felt nervous before a big decision.
Wait, that reminds me of something hilarious. Didn't you find examples of AI hosts trying to be relatable?
Oh man, yes, I found these AI hosts saying things like, sometimes I just feel like being lazy and I've definitely procrastinated myself.
(09:05):
It's like, buddy, you're a computer program designed for productivity. You don't have lazy days.
That's where it gets weird and uncanny-valley-ish. But let's talk about the technical stuff. Are the voices just not good enough yet?
Actually, the voices are surprisingly good. The real problems are deeper.
AI-generated podcasts make factual errors constantly.
(09:27):
In one case, an AI podcast attributed a story to someone called Jess Mars, who turned out to be a fictional character from a different AI experiment.
No way! So the AI was basically citing its own hallucination as a source?
Exactly, and here's what's scary. People tend to trust what computer programs tell them, even when they're wrong.
(09:48):
But when that false information comes from a friendly, authoritative voice, it becomes way more convincing.
That's actually terrifying from a misinformation perspective. But I'm curious about the content quality.
Even if the facts were right, does AI-generated content just feel... generic?
One researcher listened to 200 AI-generated podcasts. I know, dedication to science.
(10:11):
And she said the AI developed this obsession with the phrase, deep dive.
Every conversation was about taking a deep dive into something.
Oh no, they caught corporate speak syndrome.
Right, and the banter became completely formulaic. It was impressive at first.
But after a few episodes, you could predict exactly how the conversations would flow.
There's no spontaneity, no real chemistry between the hosts.
(10:34):
So even when the AI gets the technical aspects right, it's missing that human unpredictability that makes conversations engaging.
But here's where it gets really interesting. This resistance to AI-generated content extends beyond just hosting.
When researchers looked at podcast discovery, they found that algorithms only account for 3% of how people find new shows,
(10:58):
AI recommendations, 0%.
Hold up, are you telling me that in the age of Netflix algorithms and Spotify recommendations,
podcast listeners are actively avoiding AI-driven discovery?
That's exactly what I'm telling you. People overwhelmingly prefer recommendations from humans they trust.
They'd rather type a topic into a search bar and do their own research than trust an algorithm to suggest what they should listen to.
(11:22):
This is blowing my mind because it goes against everything we assume about how people interact with AI.
What do you think is driving this resistance?
I think it comes down to authenticity.
Merriam-Webster even named authenticity, their word of the year in 2023, reflecting this cultural moment where people are desperately seeking what's genuinely real.
(11:44):
And podcasting has always been about that intimate, authentic connection.
It's like the difference between getting advice from a trusted friend versus reading a chatbot's response to your problems.
Exactly, and the industry is responding.
Podmatch, which connects podcast hosts with guests, has started adding features specifically to keep AI out of the equation.
(12:06):
They want to ensure all interactions remain human.
So is this just a temporary thing while the technology improves, or are we seeing something more fundamental about human nature?
I think it's fundamental.
The research suggests that people readily embrace AI that amplifies human capabilities,
like editing tools that make podcasters sound more professional, or transcription services that save hours of work.
(12:29):
Right. So AI is the behind-the-scenes production assistant, not the star of the show.
Exactly, and about 40% of podcasters already use AI tools, and those tools are reducing production costs by up to 50%,
but they're using AI to enhance their human creativity, not replace it.
This actually connects to something bigger about how we're thinking about AI integration across industries, doesn't it?
(12:53):
Absolutely, we're seeing this pattern everywhere.
People love AI that makes them better at being human, but they resist AI that tries to replace human connection and creativity.
So for anyone listening who's thinking about starting a podcast or creating content, what's the takeaway here?
Don't try to replace yourself with AI, but absolutely use AI to amplify what makes you uniquely you.
(13:16):
Use it for editing, for research, for generating ideas.
But your voice, your experiences, your authentic perspective, that's irreplaceable.
And for listeners, this is actually encouraging news.
In a world where AI is automating everything, podcasting might be one of the last bastions of genuine human connection in media.
Which makes the medium even more valuable.
(13:39):
When you find a podcast host you connect with, you're not just consuming content, you're participating in a relationship that AI fundamentally cannot replicate.
So next time you're choosing between a perfectly polished AI-generated summary and that slightly rambling human host who occasionally forgets what they were saying,
choose the human. That's where the real connection happens.
(14:01):
Because at the end of the day, we don't just want information, we want understanding, empathy and authentic connection.
And that's still uniquely human.
Yaakov, thanks for going down this rabbit hole and bringing back such fascinating insights.
Always a pleasure, Alex.
That's Innovation Pulse. Keep questioning, keep connecting, and we'll catch you next time.
(14:31):
Stay tuned for more updates.
(15:01):
Thank you.
Welcome to Innovation Pulse, your quick no-nonsense update on the latest in AI.
(00:10):
First, we will cover the latest news.
Open AI is eyeing a $500 billion valuation with stock sales and new model releases,
while Claude Opus, 4.1, and Google's Gemini 2.5 are pushing coding boundaries.
After this, we will dive deep into the fascinating evolution of AI in podcasting,
(00:32):
examining how technology enhances creativity without replacing the human touch.
Open AI, the company behind ChatGPT, is considering a stock sale that could value it at around $500 billion,
up from its current valuation of $300 billion.
This sale would allow employees to cash out, highlighting Open AI's rapid growth in users and revenue,
(00:57):
driven by its flagship ChatGPT product.
Open AI's revenue has doubled in the first seven months of the year, with projections to hit $20 billion by year-end.
The company has seen a rise to 700 million weekly active users.
The stock sale discussions follow a funding round led by Softbank, aiming to raise $40 billion.
(01:22):
Tech companies are fiercely competing for AI talent, with meta-investing heavily in scale AI.
Open AI may shift from its capped profit model, potentially setting the stage for an IPO in the future.
Open AI has unveiled two new open-weight language models,
(01:43):
GPT-OSS, 120B and GPT-OSS, 20B, marking the first release of such models since GPT-2.
In 2019, these models are cost-effective, customizable, and designed for developers and researchers.
Unlike open-source models, open-weight models share their parameters publicly,
(02:06):
offering transparency but not complete source code access.
The release, delayed for additional safety testing, includes measures to filter harmful data and prevent malicious fine-tuning.
Open AI collaborated with companies like NVIDIA and AMD to ensure compatibility across various hardware.
(02:27):
The models available on platforms like Huggingface and GitHub can be used on consumer devices or cloud services.
Open AI aims to democratize AI access, allowing users to employ the models for tasks like advanced reasoning and personal assistance.
(02:48):
Join us as we step into the next evolution of coding excellence.
Claude Opus 4.1 is now available, enhancing real-world coding, reasoning, and agentic tasks.
It's accessible to paid Claude users and via Claude code, with no change in pricing from Opus 4.
The upgrade is available through the API, Amazon Bedrock, and Google Cloud's Vertex AI.
(03:14):
Opus 4.1 achieves a 74.5% coding performance on the SWE bench verified and improves research and data analysis skills.
Focusing on detail tracking and agentic search, GitHub highlights notable advancements in multi-file code refactoring, while Raccoon Group praises its precision in large code-based corrections.
(03:39):
WinSurf reports a significant improvement, comparable to past model upgrades.
Developers are encouraged to upgrade to Opus 4.1, accessible via Claude Opus 4.1, 2020 and 250805.
Feedback is welcome to aid in developing future models.
(04:02):
Google has introduced its most advanced AI model, Gemini 2.5 DeepThink, but access is limited to those subscribing to the $250 AI Ultra plan.
Released in the Gemini app, DeepThink is designed for complex queries, utilizing more computing resources than other models.
It enhances the foundation of Gemini 2.5 Pro by increasing thinking time and employing greater parallel analysis.
(04:31):
This allows it to explore multiple problem-solving approaches, revisiting and remixing hypotheses for higher quality outputs.
Although response times are longer, DeepThink excels in design aesthetics, scientific reasoning and coding.
It has undergone standard benchmarks, outperforming Gemini 2.5 Pro, and rivals like OpenAI 03 and Grock 4.
(04:57):
In a challenge called Humanity's Last Exam involving 2,500 complex questions across 100 subjects, DeepThink achieved a score of 34.8%, significantly surpassing other models.
11 Labs has unveiled a new AI model that generates music cleared for commercial use, expanding its focus beyond AI audio tools.
(05:24):
Known for its text-to-speech products, 11 Labs is now venturing into music, offering samples like a synthetic voice, rapping in a style reminiscent of artists like Dr. Dre and Kendrick Lamar.
This raises concerns about the source material for AI music training, as seen in lawsuits against Suno and Udio by the Recording Industry Association of America for using copyrighted music.
(05:50):
11 Labs has partnered with Merlin Network and Cobalt Music Group to license music for AI training. Cobalt ensures artists must opt in, offering benefits like new revenue streams, revenue sharing and safeguards against misuse, according to a Cobalt representative.
And now, pivot our discussion towards the main AI topic.
(06:20):
Hey everyone, Alex here with Innovation Pulse, and today I've got Yakov Lasker with me, who just spent way too much time diving into one of the strangest contradictions in tech right now.
Yakov, before we even say what we're talking about, hit me with that number that made my brain hurt.
Okay, so picture this. The AI podcasting market is exploding at 28% growth per year, heading toward $26 billion by 2033.
(06:48):
Meanwhile, when researchers actually tested human versus AI-generated podcasts, 96% of people chose the human version. 96%, Alex.
Wait, hold up. So we're throwing billions at AI podcast technology that people actively don't want?
Exactly, and here's the kicker. About 45 million Americans listen to AI-driven podcasts every month. But when you dig deeper, they're not actually listening to robot hosts. They're listening to humans who use AI tools for editing and production.
(07:18):
Oh, that's like saying people love AI music because they stream songs that were auto-tuned. The AI is behind the scenes, not replacing the artist.
Bingo, and this disconnect is teaching us something huge about how humans really interact with technology, especially when it comes to storytelling and connection.
Okay, so before we unpack why people are rejecting AI hosts, let's set the stage here. This isn't just about podcasting, right? This is happening while AI is taking over everything from writing to art to customer service.
(07:50):
Right, and podcasting seemed like the perfect candidate for AI takeover. I mean, it's just talking, right? We've got voice synthesis that can clone your voice from 10 seconds of audio.
Google's Notebook LM can turn any document into a conversation between two hosts that sounds incredibly realistic.
I've heard those Notebook LM podcasts. They're honestly creepy how good they sound, but you're telling me people still aren't buying it?
(08:14):
Here's where it gets fascinating. When researchers dug into the psychology of podcast listening, they found that podcasting succeeds because it creates what they call parasocial relationships.
Listeners feel like they actually know the hosts, like they're friends hanging out.
Oh, that makes total sense. I've definitely caught myself thinking, I wonder what Joe Rogan would say about this during random moments in my day.
(08:39):
Exactly. And here's the thing, AI can't authentically participate in that relationship because it has no genuine experiences to share.
It can't tell you about the time it bombed a presentation or felt nervous before a big decision.
Wait, that reminds me of something hilarious. Didn't you find examples of AI hosts trying to be relatable?
Oh man, yes, I found these AI hosts saying things like, sometimes I just feel like being lazy and I've definitely procrastinated myself.
(09:05):
It's like, buddy, you're a computer program designed for productivity. You don't have lazy days.
That's where it gets weird and uncanny-valley-ish. But let's talk about the technical stuff. Are the voices just not good enough yet?
Actually, the voices are surprisingly good. The real problems are deeper.
AI-generated podcasts make factual errors constantly.
(09:27):
In one case, an AI podcast attributed a story to someone called Jess Mars, who turned out to be a fictional character from a different AI experiment.
No way! So the AI was basically citing its own hallucination as a source?
Exactly, and here's what's scary. People tend to trust what computer programs tell them, even when they're wrong.
(09:48):
But when that false information comes from a friendly, authoritative voice, it becomes way more convincing.
That's actually terrifying from a misinformation perspective. But I'm curious about the content quality.
Even if the facts were right, does AI-generated content just feel... generic?
One researcher listened to 200 AI-generated podcasts. I know, dedication to science.
(10:11):
And she said the AI developed this obsession with the phrase, deep dive.
Every conversation was about taking a deep dive into something.
Oh no, they caught corporate speak syndrome.
Right, and the banter became completely formulaic. It was impressive at first.
But after a few episodes, you could predict exactly how the conversations would flow.
There's no spontaneity, no real chemistry between the hosts.
(10:34):
So even when the AI gets the technical aspects right, it's missing that human unpredictability that makes conversations engaging.
But here's where it gets really interesting. This resistance to AI-generated content extends beyond just hosting.
When researchers looked at podcast discovery, they found that algorithms only account for 3% of how people find new shows,
(10:58):
AI recommendations, 0%.
Hold up, are you telling me that in the age of Netflix algorithms and Spotify recommendations,
podcast listeners are actively avoiding AI-driven discovery?
That's exactly what I'm telling you. People overwhelmingly prefer recommendations from humans they trust.
They'd rather type a topic into a search bar and do their own research than trust an algorithm to suggest what they should listen to.
(11:22):
This is blowing my mind because it goes against everything we assume about how people interact with AI.
What do you think is driving this resistance?
I think it comes down to authenticity.
Merriam-Webster even named authenticity, their word of the year in 2023, reflecting this cultural moment where people are desperately seeking what's genuinely real.
(11:44):
And podcasting has always been about that intimate, authentic connection.
It's like the difference between getting advice from a trusted friend versus reading a chatbot's response to your problems.
Exactly, and the industry is responding.
Podmatch, which connects podcast hosts with guests, has started adding features specifically to keep AI out of the equation.
(12:06):
They want to ensure all interactions remain human.
So is this just a temporary thing while the technology improves, or are we seeing something more fundamental about human nature?
I think it's fundamental.
The research suggests that people readily embrace AI that amplifies human capabilities,
like editing tools that make podcasters sound more professional, or transcription services that save hours of work.
(12:29):
Right. So AI is the behind-the-scenes production assistant, not the star of the show.
Exactly, and about 40% of podcasters already use AI tools, and those tools are reducing production costs by up to 50%,
but they're using AI to enhance their human creativity, not replace it.
This actually connects to something bigger about how we're thinking about AI integration across industries, doesn't it?
(12:53):
Absolutely, we're seeing this pattern everywhere.
People love AI that makes them better at being human, but they resist AI that tries to replace human connection and creativity.
So for anyone listening who's thinking about starting a podcast or creating content, what's the takeaway here?
Don't try to replace yourself with AI, but absolutely use AI to amplify what makes you uniquely you.
(13:16):
Use it for editing, for research, for generating ideas.
But your voice, your experiences, your authentic perspective, that's irreplaceable.
And for listeners, this is actually encouraging news.
In a world where AI is automating everything, podcasting might be one of the last bastions of genuine human connection in media.
Which makes the medium even more valuable.
(13:39):
When you find a podcast host you connect with, you're not just consuming content, you're participating in a relationship that AI fundamentally cannot replicate.
So next time you're choosing between a perfectly polished AI-generated summary and that slightly rambling human host who occasionally forgets what they were saying,
choose the human. That's where the real connection happens.
(14:01):
Because at the end of the day, we don't just want information, we want understanding, empathy and authentic connection.
And that's still uniquely human.
Yaakov, thanks for going down this rabbit hole and bringing back such fascinating insights.
Always a pleasure, Alex.
That's Innovation Pulse. Keep questioning, keep connecting, and we'll catch you next time.
(14:31):
Stay tuned for more updates.
(15:01):
Thank you.
Popular Podcasts
New Heights with Jason & Travis Kelce
Football’s funniest family duo — Jason Kelce of the Philadelphia Eagles and Travis Kelce of the Kansas City Chiefs — team up to provide next-level access to life in the league as it unfolds. The two brothers and Super Bowl champions drop weekly insights about the weekly slate of games and share their INSIDE perspectives on trending NFL news and sports headlines. They also endlessly rag on each other as brothers do, chat the latest in pop culture and welcome some very popular and well-known friends to chat with them. Check out new episodes every Wednesday. Follow New Heights on the Wondery App, YouTube or wherever you get your podcasts. You can listen to new episodes early and ad-free, and get exclusive content on Wondery+. Join Wondery+ in the Wondery App, Apple Podcasts or Spotify. And join our new membership for a unique fan experience by going to the New Heights YouTube channel now!
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!