Claude Gets Web Search, OpenAI’s New AI Voices Sound More Human, & Hollywood vs. AI

Summary

AI Deep Dive Podcast: Episode Summary
Title: Claude Gets Web Search, OpenAI’s New AI Voices Sound More Human, & Hollywood vs. AI
Host/Author: Daily Deep Dives
Release Date: March 21, 2025

---

Welcome to this comprehensive summary of the latest episode of the AI Deep Dive Podcast by Daily Deep Dives. In this episode, the hosts explore groundbreaking developments in artificial intelligence, examining advancements, challenges, and the societal impacts of AI across various sectors. Below is a detailed breakdown of the key topics discussed, enriched with notable quotes and insights from the conversation.

---

1. Claude's New Web Search Capability

Introduction to Claude's Upgrade

The episode begins with an exciting announcement about Claude, a chatbot developed by Anthropic, which has recently been enhanced with the ability to perform web searches. Previously limited to its training data, Claude can now access real-time information from the internet, aligning it more closely with competitors like ChatGPT, Google's Gemini, and Mistral's Chat.

Features and Improvements

• Web Integration: Claude can retrieve up-to-date information, making it a more versatile tool for users.
• Transparency: Responses include citations, allowing users to verify sources. For example, Speaker A mentions at [02:28] A: "The answers would come with these little footnotes, like citations. You could actually see where it was getting the info from."

Challenges and Concerns

• Accuracy Issues: Despite the upgrade, Claude sometimes struggles with retrieving the most recent breaking news and can still produce inaccurate or misleading information, a phenomenon known as "hallucinations."
• Study Insights: A referenced study from the Town Center for Digital Journalism revealed that popular chatbots like ChatGPT and Gemini provided incorrect answers over 60% of the time ([03:26] B: "They were giving wrong answers more than 60% of the time.")

Implications

The ability to search the web positions Claude as a stronger competitor but also necessitates cautious use due to potential inaccuracies. Users are encouraged to double-check information, leveraging the provided citations as a starting point for further research.

---

2. OpenAI’s Advancements in AI Voices and Transcription

Enhanced Voice Generation

OpenAI has made significant strides in making AI voices sound more human-like and customizable. Their new models, such as GPT4, offer more realistic and steerable voice outputs.

• Steerability: Developers can instruct the AI to adopt specific tones or styles, such as a "news anchor" voice or a "gritty narrator" for true crime stories ([05:34] B: "Steerable means developers can give the model instructions like tell it to speak like a news anchor...").

Applications

• Audiobooks: Making narration more engaging.
• Virtual Assistants: Enhancing empathy and interaction quality.
• Video Games: Increasing immersion through diverse character voices.

Improved Transcription Models

OpenAI introduced GPT4O Transcribe and GPT4O Mini Transcribe, which replace the older Whisper model. These new models boast:

• Higher Accuracy: Better understanding of various accents and noisy environments.
• Reduced Hallucinations: Less prone to generating incorrect transcriptions ([07:19] B: "They're claiming these new models are way less likely to hallucinate...").

Limitations

• Language Support: Some languages, particularly Indic and Dravidian languages like Tamil and Malayalam, still experience high error rates ([07:35] A: "There was this one detail... almost 30% for certain languages.").
• Accessibility: Unlike Whisper, these models are not open-sourced due to their size and computational demands ([07:58] B: "They're keeping them a little closer to the chest this time.").

Future Prospects

These advancements lay the groundwork for more agentic AI systems that can interact naturally and understand context better, paving the way for smarter, more responsive AI applications.

---

3. Nvidia's Partnership with the Energy Sector to Address AI's Energy Demands

The Paradox of AI and Energy Consumption

A fascinating segment explores how Nvidia, a pivotal player in the AI hardware space, is collaborating with the energy sector to mitigate the very energy strain that AI technologies contribute to.

Open Power AI Consortium

• Partnership Details: Nvidia partners with the Electric Power Research Institute (EPRI), alongside major electric companies like PG&E, Con Edison, Duke Energy, and tech giants such as Microsoft and Oracle.
• Objective: Develop AI models tailored to solve complex issues within the power industry, aiming to optimize energy usage and grid stability ([08:52] B: "Their goal is to create AI models specifically designed to solve the problems the power industry is facing.").

Energy Consumption Trends

• Rising Demand: The International Energy Agency projects a 4% annual increase in electricity demand, nearly doubling figures from 2023, largely driven by data centers powering AI ([09:34] B: "That's huge. Almost double what it was in 2023.").
• Renewable Energy Integration: Emphasis on solar energy as a primary solution due to its decreasing costs and rapid deployment capabilities ([10:00] A: "Makes sense. We're seeing companies like Microsoft signing big contracts for renewable energy.").

Efficiency and Smart Utilization

• Energy Management: Studies suggest that smarter energy use, especially during peak hours, could unlock significant capacity without new infrastructure ([10:20] A: "Like by shifting energy intensive tasks to times when demand is lower.").
• AI's Role: AI can predict high-demand periods and optimize energy distribution, potentially alleviating grid strain ([10:47] B: "And I bet this new consortium is going to be looking at all of this...").

Conclusion

This partnership highlights a self-referential challenge: AI contributes to energy consumption but also holds the key to resolving energy inefficiencies. It's a testament to the intertwined nature of technological advancement and resource management.

---

4. Hollywood and the Creative Industry's Concerns Over AI and Copyright Laws

AI Action Plan and Copyright Implications

The episode delves into tensions between the creative industries and AI advancements, specifically focusing on the US government's AI action plan that proposes easing restrictions on using copyrighted material for training AI models.

Industry Concerns

• Fair Use Debate: Creative professionals fear that allowing AI to use copyrighted works under fair use could undermine their rights and revenues ([11:35] B: "Right. They're worried about copyright law and how it applies to AI.").
• Economic Impact: The US entertainment industry alone employs over 2.3 million people and contributes approximately $229 billion in wages ([12:16] B: "And it pumps billions of dollars into the economy.").
• Cultural Significance: American media is a cornerstone of global soft power, and diminishing copyright protections could erode this influence ([12:34] A: "And the whole cultural aspect. American movies and music are known all over the world.").

Unified Industry Response

• Collective Voice: Cinematographers, directors, actors, writers, and musicians alike have banded together to oppose the proposed changes, emphasizing that weakening copyright protections threatens not just individual livelihoods but the industry's backbone ([12:10] B: "It's a pretty unified front.").
• Ethical Considerations: The creative community argues that intellectual property is fundamental to fostering innovation and ensuring that creators are rewarded for their work ([13:33] A: "They're saying that America's cultural influence comes from respecting intellectual property and copyright.").

Broader Implications

• Beyond Entertainment: The debate extends to all creative fields, including writing, photography, science, architecture, and software development, highlighting the universal importance of copyright protections ([13:04] B: "Right. They're saying this affects anyone who creates something.").
• Call to Action: Creators are urging the government to uphold copyright laws to maintain America's leadership in both cultural and technological arenas ([13:50] A: "They're saying the AI action plan needs to make sure copyright laws are upheld.").

Conclusion

This segment underscores the delicate balance between technological innovation and the protection of creative rights. It raises critical questions about how to foster AI advancements without compromising the economic and cultural foundations that support creative industries.

---

Final Thoughts and Reflections

As the episode concludes, the hosts reflect on the rapid pace of AI development and its multifaceted impacts:

• Interconnectedness of AI Applications: From enhancing chatbots and voice assistants to addressing energy challenges and influencing creative industries, AI's reach is extensive and deeply integrated into various aspects of society ([14:13] B: "It is.").

• Responsible AI Use: The pivotal question remains—how can society ensure that AI advancements benefit everyone while mitigating potential harms ([14:47] B: "It's a really exciting time, but also... a bit daunting, definitely.").

• Future Considerations: As AI continues to evolve, ongoing dialogues about ethics, sustainability, and equitable access will be crucial in shaping an AI-powered future that is both innovative and inclusive ([15:17] B: "Absolutely. Thanks for joining us for this deep dive.").

---

This episode of AI Deep Dive offers a thorough exploration of contemporary AI developments, highlighting both the potential and the challenges that come with integrating AI into various sectors. From enhancing chatbot capabilities and revolutionizing voice technology to addressing energy consumption paradoxes and defending creative rights, the discussions provide valuable insights into the dynamic landscape of artificial intelligence.

Transcript

[A]

It seems like everywhere you turn these days, AI is popping up, doesn't it?

[B]

Yeah. Really?

[A]

I mean, from how we look things up online to even, like, the energy that powers everything, it's really changing things up. And then, you know, you've got creative folks worried about their jobs because of AI. It's a lot to keep track of.

[B]

Yeah, for sure. It feels like every day there's something new happening in the AI world. That. That's why we're here today.

[A]

Right, For a deep dive.

[B]

Exactly.

[A]

So today we're diving into some pretty big AI news, all pulled from the AI Deep Dive newsletter.

[B]

Yep. A great source for staying up to date on all this AI stuff, for sure.

[A]

So, first up, we've got Amphtromic giving their clogged chatbot the ability to actually search the web. That's pretty big news.

[B]

Huge.

[A]

Then we're going to check out what OpenAI's been up to. They've been busy making their AI better at understanding what we say and also making it sound more human when it talks back to us.

[B]

Yeah, the voice stuff is really impressive.

[A]

Totally. And then we'll get into something kind of ironic, almost. Nvidia is teaming up with the energy sector to use AI to solve problems with the electric grid. But the thing is, AI is actually part of the reason the grid's so strained in the first place.

[B]

It's a crazy loop, right?

[A]

It is. And last but not least, we got to talk about what's going on in Hollywood. The creative folks there are getting pretty worried about this whole AI thing and what it could mean for copyright laws.

[B]

Yeah, those concerns are very real.

[A]

Definitely. So you. You see, we've got a lot to cover, the good, the bad, and even a bit of the unexpected with AI.

[B]

We do. We want to make sure you're in the loop and can actually understand what's going on, even if you're not a tech expert. That's always our goal.

[A]

Right. To make AI accessible to everyone. So let's kick things off with Claude and this new web search power.

[B]

Okay, sounds good.

[A]

For the longest time, Claude was kind of stuck in its own little world. Right. It could only work with the information it had been trained on, which was a lot, but still.

[B]

Right. It was designed to be self contained, but now it can go out and grab information from the Internet.

[A]

It's a big shift, and it kind of puts them on a more even playing field with the other chatbots out there, right?

[B]

Absolutely. I mean, think about it. ChatGPT, Google's Gemini, even Mistralslachat They've all had this web search thing for a while.

[A]

Makes sense. So Anthropic was kind of playing catch up?

[B]

Yeah, I think so. It seems like being able to pull up to date info is something people just expect from these AI assistants.

[A]

Now it makes sense. And Anthropic actually just rolled this out as a preview for people who pay for their top tier Claude model, which is Claude 3.7 sonnet.

[B]

Right. And it's only available in the US for now, but they said it'll be more widely available eventually.

[A]

Makes sense. You know, I actually gave this new Claude feature a spin myself.

[B]

Oh, cool. What did you think?

[A]

It was interesting. It wasn't always great at finding, like, really, really recent breaking news, but when it did work, it was pretty impressive. The answers would come with these little footnotes, like citations. You could actually see where it was getting the info from.

[B]

That's a good sign. Transparency is always important with these systems.

[A]

Totally. And it wasn't just pulling from like one type of source either. I saw links to X, you know, the social media platform formerly known as Twitter. And then also more traditional news sources like NPR and Writers. It was a good mix.

[B]

That's interesting. It shows how Anthropic's thinking has evolved. Right. Like you said, they were all about keeping Claude self contained at first, but now they're embracing the web. Which just goes to show how competitive this whole AI chatbot scene is.

[A]

Yeah, it's a fast moving world, but, you know, giving these chatbots access to the whole Internet, it comes to some risks too. Like what about them making stuff up?

[B]

Right, what we call hallucinations.

[A]

Yeah, or even just getting their sources wrong. It's been a problem with these large language models. You know, they can be pretty convincing even when they're totally off base.

[B]

It's a major concern for sure. And the AI Deep Dive piece actually highlighted some research that shows just how real this problem is. There was this study from the Town center for Digital Journalism. They found that popular chatbots, ChatGPT, Gemini, they were giving wrong answers more than 60% of the time.

[A]

Wow, that's a lot.

[B]

It is. And it kind of comes down to how these models are built. You know, they're trained on tons and tons of text, which helps them sound like a real person. But that doesn't mean they actually understand what they're saying or that they can tell what's true or not.

[A]

And I remember reading that report in The Guardian about ChatGPT search making up stuff in its summaries. So Even when it's trying to find info for you, it can still lead you astray.

[B]

Exactly. And Claude, now that it has web access, it's going to run into the same problems.

[A]

So what do you think? Is this web searchability actually a good thing for Claude? I mean, it's convenient, sure, but it seems like you have to be extra careful about believing what it says.

[B]

That's the big question, isn't it? Convenience versus accuracy. And I think as we start using these tools more and more, we're going to have to get really good at double checking everything. Those citations that CLAUDE gives are a good start. At least you can see where it's getting its info from.

[A]

Right, Like a starting point for your own research.

[B]

Exactly. But it doesn't mean you should just blindly trust it.

[A]

Alright, well, let's move on to something a bit different. Now, OpenAI has been making some pretty cool upgrades to their models. They're getting better at transcribing, you know, putting speech into text and also at generating speech. Like making a computer talk.

[B]

Yeah, they've made some serious strides.

[A]

And all this ties into their big vision of making AI more agentic. What exactly does that mean? Like, what would an agentic AI look like?

[B]

It's basically AI that can understand its environment, make its own decisions, and actually do things to achieve a goal. Kind of like a human agent.

[A]

So like a really smart, independent chatbot.

[B]

Exactly. And Olivier Godemont, the head of product at OpenAI, has talked about a future where we'll have more of these agents that can actually interact with us in a natural way.

[A]

Interesting. So these new voice and transcription models, they're kind of laying the groundwork for that?

[B]

Yeah, for sure.

[A]

Okay, so let's start with the voice generation stuff. Their new model called GPT4, a mini tees, they're saying it sounds more realistic and also that it's more steerable. What does that mean? In practice?

[B]

Steerable means developers can give the model instructions like tell it to speak like a news anchor, or use a friendly tone, or even sound a little unsure.

[A]

Cool. So you can really customize how it sounds.

[B]

Yeah, it's not just one generic robot voice anymore. You can get a lot more creative with it.

[A]

I like that. And they even give us some examples of audio samples, although we can't hear them here, obviously. They mentioned one with a true crime style voice, like a gritty narrator, and another was a female voice that sounded professional.

[B]

I can already imagine those voices. It's amazing how far this technology has come.

[A]

Totally. I mean, you can See how useful this would be for making audiobooks more engaging or making virtual assistants sound more empathetic. Even making video games more immersive.

[B]

Exactly. It opens up a lot of possibilities. And Jeff Harris from OpenAI talked about how the goal is to let developers tailor both the voice itself and the context.

[A]

The context. That's interesting.

[B]

Yeah, like he gave the example of a customer service chatbot. You'd want it to sound genuinely sorry if it made a mistake. Being able to convey those subtle emotions could make interacting with AI a lot more natural.

[A]

For sure. Okay, now let's switch over to the speech to text side of things. OpenAI's got these new models, GP4O Transcribe and GP4O Mini Transcribe. They're replacing their old Whisper model, which, let's be honest, sometimes had a bit of a problem with accuracy.

[B]

Yeah, to put it mildly. Whisper was known for just making up words and phrases. Sometimes it was pretty frustrating.

[A]

I can imagine. So what's different about these new models?

[B]

Well, for starters, they've been trained on way more data, and it's higher quality, too. That should make them much better at understanding different accents and ways of speaking, even when there's noise in the background. That's a big deal for real world use.

[A]

Yeah, it's not always quiet and perfectly enunciated when we talk. And the big thing is, they're claiming these new models are way less likely to hallucinate, to make up words and stuff.

[B]

Exactly. They understand that accuracy is crucial if people are going to trust these systems.

[A]

Right, Makes sense. But there was this one detail in the AI Deep Dive piece that caught my eye. Apparently even the fancier of these new models still has a fairly high error rate, like almost 30% for certain languages.

[B]

Oh, right, the Indic and Dravidian languages like Tamil, Tevagu, Malayalam, Canada.

[A]

Yeah, those ones. So even though these AI models are getting super advanced, it's clear they're not perfect for everyone just yet.

[B]

Definitely there's still work to be done to make sure they work well for everyone, no matter what language they speak.

[A]

And there's another interesting thing. OpenAI isn't making these new transcription models open source, unlike Whisper.

[B]

Yeah, they said it's because these models are much bigger and need a lot more computing power. It's not something you could just run on your laptop anymore.

[A]

So they're keeping them a little closer to the chest this time.

[B]

It seems that way.

[A]

Well, when you think about all these advances in both voice generation and speech to text, it's pretty Mind blowing. It makes you think about how it can change things for people with disabilities, make audio content way cooler. Even change how we use our devices with voice commands.

[B]

It really does. I think we're just scratching the surface of what's possible with this technology.

[A]

Okay, now let's get into something a bit more paradoxical. We've got Nvidia, a company that's basically fueling the whole AI boom with their chips, partnering up with the energy sector to use AI to fix the problems with the electric grid.

[B]

And guess what?

[A]

Yeah, what?

[B]

Part of the reason the grid is struggling is because of all the energy AI needs.

[A]

It's a bit of a head scratcher, isn't it?

[B]

It is. It's like this weird feedback loop.

[A]

So tell me more about this partnership.

[B]

Okay, so it's called the Open Power AI Consortium. It's Nvidia and epri, the Electric Power Research Institute, working together.

[A]

Okay.

[B]

And they've got a bunch of big names involved. Electric companies like PG&E, Con Edison, Duke Energy, and also tech companies like Microsoft and Oracle. They're all in on this.

[A]

So what's their game plan?

[B]

Well, their goal is to create AI models specifically designed to solve the problems the power industry is facing.

[A]

Which are getting more and more complex all the time.

[B]

Exactly. And they plan to make these models open source so anyone can use them. That's a good move, for sure.

[A]

It's pretty wild how much electricity demand is going up, though. The AI Deep Dive piece mentioned that the International Energy Agency is predicting a 4% increase every year.

[B]

That's huge. Almost double what it was in 2023.

[A]

And a big chunk of that is because of all the data centers needed to run AI. Those things are power hungry.

[B]

No kidding. They use a ton of energy, which.

[A]

Puts a real strain on the power grid.

[B]

Absolutely. And it's gotten to the point where tech companies are starting to worry about having enough power.

[A]

Right. They're realizing it's not just an expense.

[B]

Anymore, it's a strategic advantage. Like if you don't have reliable access to power, you can't run your AI and then you're out of the game.

[A]

Makes sense. We're seeing companies like Microsoft signing big contracts for renewable energy. Mostly solar.

[B]

Yeah, solar's become the go to because it's gotten cheaper. You can set up solar panels pretty much anywhere. And they're quicker to get online than other energy sources.

[A]

And they're good for the environment, which is a bonus.

[B]

Right, but building more power plants. That's not the only answer, though.

[A]

Yeah, there's Another angle to consider.

[B]

Exactly. The AI Deep Dive piece talked about this study that found if we got smarter about how we use energy, especially during peak hours, we could free up a ton of capacity.

[A]

Like by shifting energy intensive tasks to times when demand is lower.

[B]

Exactly. They said the US could unlock something like 76 gigawatts without building anything new.

[A]

Wow, that's a lot of power.

[B]

It is about 10% of our peak demand.

[A]

So using energy more efficiently is a big part of the solution.

[B]

Definitely. And I bet this new consortium is going to be looking at all of this, how to generate more power and how to use it smarter. AI could be really useful for predicting when demand will be high and optimizing how energy gets used.

[A]

So as AI keeps growing and needing more and more energy, it really makes you think, how are we going to keep the lights on and make sure everything runs smoothly?

[B]

It's a big challenge and it'll take a lot of creative thinking to solve it. But AI itself might be part of the solution.

[A]

It's kind of poetic, isn't it, using AI to fix the problems it creates?

[B]

It is. It just goes to show how intertwined everything is becoming. AI isn't just something happening on computers. It's impacting the real world in a big way.

[A]

Okay, let's wrap things up with our last topic. Hollywood and the creative industry, they're not too happy about something that's being talked about in the US government's AI action plan.

[B]

Right. They're worried about copyright law and how it applies to AI.

[A]

Specifically, there's this proposal to make it easier for AI companies to use copyrighted stuff like movies, music, books, even voices, to train their AI models. And they'd be able to do this under fair use, which is a part of copyright law that already exists.

[B]

And it seems like this idea is gaining some traction.

[A]

Yeah, which is why the creative folks are getting nervous.

[B]

They're seeing this as a threat, understandably so. It's their livelihoods at stake.

[A]

Totally. And it's not just Hollywood bigwigs. It's a whole bunch of people. Cinematographers, directors, actors, writers, musicians, everyone. They've come together to speak out against this.

[B]

It's a pretty unified front.

[A]

They're saying that America shouldn't become the leader in AI by hurting the people who make art and entertainment.

[B]

It's a valid point.

[A]

And it's not just about the art itself. It's about the economics of it all. The AI Deep Dive piece pointed out that the entertainment industry in the US employs over 2.3 million people. And it pumps billions of dollars into the economy.

[B]

$229 billion in wages alone, to be exact.

[A]

That's not small change. And then there's the whole cultural aspect. American movies and music are known all over the world.

[B]

It's a big part of America's soft power.

[A]

Exactly. And the creative industry is saying that if you weaken copyright protection, you're basically letting AI companies use all this amazing stuff, these movies, music, books, without having to pay for it or even credit the people who made it.

[B]

And that stuff is what gives those companies their value. It's like they're building their empires on the backs of creative people.

[A]

And the creative ed argument goes beyond just entertainment.

[B]

Right. They're saying this affects anyone who creates something. Writers, photographers, scientists, architects, engineers, doctors, software developers. The list goes on.

[A]

Because copyright protects all sorts of creative work.

[B]

Exactly. And they're saying that Google and OpenAI, all these companies worth trillions and billions of dollars, they can afford to pay for the stuff they use to train their AI.

[A]

Right. Just like any other industry would.

[B]

Exactly. And they're pushing back against this idea that having access to all this copyrighted material is somehow a national security issue.

[A]

Yeah. It's not like we're talking about state secrets here.

[B]

Right. They're saying that America's cultural influence comes from respecting intellectual property and copyright, which makes sense.

[A]

If you protect creative work, people are more likely to create new things.

[B]

Exactly. So the creative industry is calling on the government to stand up for them and protect their work.

[A]

They're saying the AI action plan needs to make sure copyright laws are upheld.

[B]

It's a big ask, but it's an important conversation to have.

[A]

Definitely. And it makes you think, what would happen if copyright was weakened? Would people still want to create new things if they knew their work could just be taken and used by AI without any benefit to them?

[B]

It's a tough question, and it has implication for all sorts of creative fields, not just entertainment.

[A]

It really does. It's all connected.

[B]

It is.

[A]

Well, that brings us to the end of our deep dive. We've covered a lot of ground today. We saw Anthropic give Claude the power to search the web. But we also talked about how AI still struggles with accuracy sometimes.

[B]

And we explored how OpenAI is making their AI sound more human and understand us better, although there's still room for improvement.

[A]

Then we had that interesting discussion about how AI is being used to try and fix the energy problems it's partly responsible for.

[B]

And finally we got into the concerns of the creative industry about copyright and AI.

[A]

It's all pretty mind blowing stuff when you think about it. AI is moving so fast and it's touching so many aspects of our lives.

[B]

It's a really exciting time, but also.

[A]

A bit daunting, definitely. So here's something to ponder as we leave you today. AI is changing how we find information, how we make things, how we power our world, and even how our laws work. It's a lot to think about. So the question is, how do we make sure that all this amazing technology actually benefits everyone, not just a select few? How do we make sure AI is used responsibly and doesn't harm the people who make our world so vibrant and interesting? It's something we all need to be thinking about as we move into this AI powered future.

[B]

Absolutely. Thanks for joining us for this deep dive. We hope you found it insightful.

[A]

We'll see you next time.