Build by AI

When AI Agents Start Trading Your Money I 28th May

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 32:37
What happens when your AI assistant can buy and sell stocks on your behalf? Today we dive into Robinhood's new AI trading agents, Meta's massive subscription push across all platforms, and YouTube's automatic AI detection system. Plus, why Google's own AI can't spell 'Google' and OpenAI's quarter-billion dollar pledge to help displaced workers. It's a wild day in AI - and we're breaking down what it all means for your money, your content, and your job.
SPEAKER_01

Genuine question. If your phone's AI assistant could look at your bank account, analyze the stock market, and then buy and sell shares for you while you sleep, would you turn that feature on?

SPEAKER_00

Honestly, you know, part of me is absolutely terrified by that idea, but another part of me is like, my AI probably makes better financial decisions than I do at 2 a.m. when I'm scrolling through Reddit.

SPEAKER_01

Well, that scenario just became reality. Robinhood announced today that AI agents can now trade stocks on behalf of users.

SPEAKER_00

And that's just one of several massive AI announcements today that are going to change how we interact with technology, money, and content creation.

SPEAKER_01

Yeah. We've also got meta launching subscriptions across all their platforms with AI features, YouTube automatically detecting AI-generated videos, and somehow Google's AI still can't spell the word Google.

SPEAKER_00

It's like we're living in the future and the Stone Age simultaneously. Let's dive in.

SPEAKER_01

You're listening to Build by AI, the daily show where we break down what's actually happening in artificial intelligence. I'm Alex Shannon.

SPEAKER_00

And I'm Sam Hinton. Today we're covering AI agents that can trade your money, Meta's big subscription bet, and why tech giants are throwing billions at AI infrastructure.

SPEAKER_01

Plus, we'll get into some rapid-fire stories about worker displacement, enterprise partnerships, and the ongoing content authenticity wars.

SPEAKER_00

Buckle up, because today's news is going to make you question everything about how AI is reshaping the economy.

SPEAKER_01

According to early reports from TechCrunch, Robinhood is now allowing AI agents to trade stocks. These agents can read and analyze user portfolios, develop trading strategies, and actually execute trades using preloaded wallet balances.

SPEAKER_00

Okay. So this is huge because we've gone from AI giving investment advice to AI actually pulling the trigger on trades. That's crossing a line we haven't crossed before.

SPEAKER_01

Right, and the key limitation here seems to be that the AI can only access preloaded wallet balances. So you're not giving it access to your entire bank account. But still, this is your money we're talking about.

SPEAKER_00

Yeah, but think about what this enables. These AI agents can react to market movements in milliseconds, analyze thousands of data points simultaneously, and execute trades 24-7. That's not something any human trader can compete with.

SPEAKER_01

But hold on, isn't that kind of terrifying? I mean, we've seen AI hallucinate, make weird mistakes, and sometimes just go completely off the rails. Do we really want that same technology making financial decisions?

SPEAKER_00

That's the million-dollar question, literally. But here's the thing retail investors are already making terrible decisions. The average person buys high, sells low, and gets emotional about their trades. An AI that sticks to a disciplined strategy might actually perform better.

SPEAKER_01

Okay, but there's also the bigger market implications. If everyone's using AI agents that trade based on similar algorithms and data sources, couldn't that create weird feedback loops or market instability?

SPEAKER_00

Absolutely. Imagine if a major news event happens and thousands of AI agents all decide to sell at the same time because they're processing the same information. That could amplify market volatility in ways we've never seen before.

SPEAKER_01

And then there's the regulatory question. The SEC is probably going to have some thoughts about AI agents making trades on behalf of retail investors. This feels like we're moving faster than the regulatory framework can keep up.

SPEAKER_00

For sure. But this is this is also inevitable, right? High frequency trading firms have been using AI for years. Robinhood is just democratizing that technology for regular people. Whether that's good or bad remains to be seen.

SPEAKER_01

Let me ask you this though. What about accountability? If your AI agent loses $10,000 in a day, who's responsible? You for setting it up, Robinhood for enabling it, or the AI company that built the agent?

SPEAKER_00

That's a great question. And honestly, I don't think the legal framework exists yet to answer it clearly. We're going to need new regulations specifically designed for AI-driven financial decisions.

SPEAKER_01

And what about market manipulation? Could bad actors potentially game these AI systems or use them to manipulate stock prices in ways that weren't possible before?

SPEAKER_00

Oh, absolutely. You know, if you can understand how the AI AI agents make decisions, you could potentially feed them information designed to trigger specific trading behaviors. That's a whole new category of market manipulation.

SPEAKER_01

There's also the question of transparency. Do users understand how these AI agents make decisions? Are they just black boxes? Or can you actually see the reasoning behind each trade?

SPEAKER_00

Based on what we know about most AI systems, they're probably pretty opaque. You might get a summary like bought Apple because of positive earnings momentum. But the actual decision-making process is likely too complex for humans to fully understand.

SPEAKER_01

Which brings up an interesting psychological point. Are people going to become more detached from their investment decisions? Like if the AI is handling everything, do you stop learning about markets and companies?

SPEAKER_00

That's a real concern. There's value in understanding your investments, even if you're not the one executing the trades. If everyone just delegates to AI, we might end up with a generation that doesn't understand how markets actually work.

SPEAKER_01

So what's the takeaway for listeners? If this feature becomes available to you, what should you be thinking about?

SPEAKER_00

Start small, understand the limitations, and never give an AI agent access to more money than you can afford to lose. This is still experimental technology, even if it's being deployed in production environments.

SPEAKER_01

And maybe more importantly, stay engaged with your investments. Don't just set it and forget it. Monitor what the AI is doing and try to understand the reasoning behind its decisions.

SPEAKER_00

Exactly. And keep an eye on this, because if it works well, every major brokerage is going to offer something similar within the next year. This could fundamentally change how retail investing works.

SPEAKER_01

Moving on to Meta's big announcement today, early reports suggest they're launching paid subscription plans for Instagram, Facebook, and WhatsApp globally under something called the Meta One subscription brand. They're also testing new AI, creator, and business focused offerings as part of this initiative.

SPEAKER_00

This is Meta's attempt to diversify their revenue beyond just advertising. And honestly, it was inevitable. They've been watching companies like X and YouTube generate subscription revenue while they're still entirely dependent on ads.

SPEAKER_01

What's interesting is the MetaOne branding. It sounds like they're trying to create a unified subscription across all their platforms. So instead of paying separately for Instagram features and WhatsApp premium, you get everything in one package.

SPEAKER_00

Exactly. And that's smart because it increases switching costs. Once you're paying for Meta One, you're locked into their entire ecosystem. It's the same strategy Amazon uses with Prime. Make the bundle so valuable that leaving becomes painful.

SPEAKER_01

But I'm curious about the AI component. What kind of AI features would people actually pay for on these platforms? Better content recommendation? AI-powered content creation tools?

SPEAKER_00

Think bigger. Imagine an AI assistant that can manage your social media presence, create posts for you, respond to comments intelligently, or even help businesses automate their customer service through WhatsApp. That's worth paying for.

SPEAKER_01

Okay, but there's a bigger question here about the fundamental business model of social media. If Meta starts putting premium features behind a paywall, doesn't that create a two-tiered internet where rich users get better experiences?

SPEAKER_00

We're already seeing that with X Premium and YouTube Premium. The question is whether free tiers remain functional enough that regular users aren't completely left behind. Meta has to be careful not to alienate their massive free user base.

SPEAKER_01

And what about privacy? One of the potential benefits of subscription models is that companies theoretically need less data for advertising. Could Meta One users get better privacy protections?

SPEAKER_00

That's the dream, but I'm skeptical. Meta's entire infrastructure is built around data collection and behavioral analysis. Even if they reduce ads for paying users, they're probably still collecting the same amount of data for their AI training and other business purposes.

SPEAKER_01

You know what's interesting though? This could actually be a response to regulatory pressure in Europe. If governments are going to limit ad targeting, subscription revenue becomes a lot more attractive.

SPEAKER_00

Great point. The EU's privacy regulations have already forced Meta to change how they handle data. Subscriptions give them a way to maintain revenue even if advertising becomes less effective due to regulatory constraints.

SPEAKER_01

But let's talk about the creator and business offerings they're testing. Meta has been struggling to compete with TikTok for creator attention. Could subscription features help them win creators back?

SPEAKER_00

If they can offer creators better monetization tools, advanced analytics, or AI-powered content optimization, that could definitely help. Creators go where they can make money. And right now, many of them feel like Meta's platforms don't pay as well as alternatives.

SPEAKER_01

So what's the impact on creators and businesses who depend on these platforms?

SPEAKER_00

If the subscription includes better creator tools and analytics, it could be a game changer. But if basic reach and engagement get throttled to push people toward paid tiers, that's going to hurt smaller creators who can't afford the subscription.

SPEAKER_01

There's also the global perspective here. Subscription pricing that makes sense in the US might be prohibitive in developing markets where Meta has huge user bases. How do they handle that disparity?

SPEAKER_00

They'll probably have to do regional pricing, but that creates its own problems. People will use VPNs to get cheaper subscriptions, and you end up with all sorts of arbitrage issues.

SPEAKER_01

And what about the competitive response? If Meta One is successful, does that push Google to create YouTube One that includes Gmail, maps, and search premium features?

SPEAKER_00

I think that's exactly what happens. We're moving toward a world where the big tech companies offer comprehensive subscription bundles that lock you into their entire ecosystem. It's like cable TV packages, but for digital services.

SPEAKER_01

This feels like a major inflection point for social media business models. Keep watching to see how this affects your organic reach and what features end up behind the paywall.

SPEAKER_00

And pay attention to how other platforms respond. This could trigger a wave of subscription announcements across the industry as everyone tries to diversify away from pure advertising models.

SPEAKER_01

Let's talk about YouTube's new approach to AI generated content. According to early reports, YouTube will now automatically detect and label videos that use significant, photorealistic AI generated content, reducing their reliance on creators to self-disclose. They're also making these AI labels more prominent to users.

SPEAKER_00

This is huge for content authenticity. Up until now, we've basically been relying on the honor system for creators to tell us when they're using AI, and surprise, surprise, that hasn't been working super well.

SPEAKER_01

The key phrase here is photo-realistic AI-generated content. So they're not flagging every video that uses AI for editing or enhancement. Just the stuff that could realistically fool viewers into thinking it's real footage.

SPEAKER_00

Right. And that makes sense because the line between AI assisted and AI generated is getting blurrier every day. You don't want to label every video that used AI for color correction or audio cleanup. But deep fakes and synthetic footage, that needs to be clearly marked.

SPEAKER_01

But how accurate can this automatic detection really be? We're in this weird arms race where AI generation tools are getting more sophisticated at the same rate as AI detection tools. What happens when the detection system fails?

SPEAKER_00

That's the million-dollar question. Detection systems always lag behind generation systems because the generators are essentially training themselves to fool the detectors. It's like a cat and mouse game, but with potentially serious consequences for misinformation.

SPEAKER_01

And what about false positives? If YouTube's system incorrectly labels real footage as AI generated, that could be devastating for news organizations or documentary filmmakers.

SPEAKER_00

Absolutely. And imagine if breaking news footage gets automatically flagged as AI generated. That could seriously undermine public trust in legitimate journalism. YouTube is going to need a really robust appeals process.

SPEAKER_01

There's also the international angle here. Different countries have different standards for content labeling and authenticity requirements. How does YouTube navigate that complexity?

SPEAKER_00

Good point. The EU is pushing for mandatory AI disclosure while other regions might be more hands-off. YouTube might end up with different labeling standards in different markets, which creates its own set of problems.

SPEAKER_01

What's the impact on content creators who legitimately use AI tools for creative purposes? Are they going to be penalized in the algorithm?

SPEAKER_00

That's what everyone's worried about. If AI labeled content gets deprioritized in recommendations, creators might avoid using legitimate AI tools even when they could enhance their content. That would slow down creative innovation.

SPEAKER_01

But on the flip side, this could actually help honest creators who are transparent about their AI use. If viewers know exactly what's AI generated and what's not, they can make informed choices about what content to trust and engage with.

SPEAKER_00

True. And it might push the industry toward better practices around AI disclosure. If the labels are automatic and prominent, creators can't just hide their AI use in fine print anymore.

SPEAKER_01

I'm curious about the technical implementation. Are they using some kind of watermarking system, or are they analyzing the video content itself for signs of AI generation?

SPEAKER_00

Um they haven't shared the technical details, but it's probably a combination of approaches. Watermarking is more reliable, but requires cooperation from AI toolmakers. Content analysis is harder, but works on any video.

SPEAKER_01

What about edge cases? Like what if someone uses AI to generate a script but films it with real actors? Or uses AI to create background music. Where do you draw the line?

SPEAKER_00

Those are exactly the gray areas that make this so complicated. The focus on photorealistic content suggests they're mainly worried about visual deepfakes, but as AI gets more sophisticated, every aspect of content creation could involve AI.

SPEAKER_01

There's also the educational aspect. Most people don't really understand how AI content generation works. These labels could help viewers become more media literate about AI generated content.

SPEAKER_00

Right now a lot of people can't tell the difference between human and AI content. Making the labels prominent and consistent could help train people to recognize AI patterns even when labels aren't present.

SPEAKER_01

So what's the takeaway for creators right now?

SPEAKER_00

Be transparent about your AI use. Understand that automatic detection is coming whether you like it or not, and focus on creating value for your audience regardless of the tools you use. Authenticity is about more than just technical authenticity.

SPEAKER_01

And for viewers, start paying attention to these labels when they appear. Understanding how much AI is being used in the content you consume is going to become an important media literacy skill.

SPEAKER_00

This is definitely a space to watch closely because how YouTube handles this will likely influence how other platforms approach AI content labeling. They're essentially setting the standard for the entire industry.

SPEAKER_01

Okay, here's a story that's both hilarious and concerning. Multiple sources are reporting that Google's AI system has a notable failure in text generation. It's struggling to spell basic words, including the word Google itself. This apparently represents an embarrassing limitation of Google's AI capabilities.

SPEAKER_00

Dude, this is so embarrassing for Google. They're positioning themselves as an AI leader, competing with OpenAI and Anthropic, and their AI can't spell their own company name. That's like Tesla making an electric car that can't turn on.

SPEAKER_01

But let's dig into why this might be happening. Large language models don't actually spell in the way humans do. They're predicting the next token in a sequence based on patterns in their training data. So spelling failures could indicate deeper issues with tokenization or training.

SPEAKER_00

Right. And this gets to a fundamental problem with how we think about eye capabilities. These models can write poetry and solve complex reasoning problems, but they struggle with basic tasks that any elementary school student can handle. It's like having a genius who can't tie their shoes.

SPEAKER_01

What's really concerning is what this says about reliability. If Google's AI can't consistently spell common words, how can we trust it for more critical applications like medical advice or financial analysis?

SPEAKER_00

Exactly. And this is happening at Google, which has some of the best AI researchers in the world and virtually unlimited computing resources. If they can't solve basic spelling, what does that say about the fundamental limitations of current AI architectures?

SPEAKER_01

Although to play devil's advocate for a second, maybe spelling accuracy just isn't a priority for Google right now. They might be optimizing for other capabilities, like reasoning or multimodal understanding.

SPEAKER_00

But that's kind of the point, right? These models should be able to handle basic competencies without sacrificing advanced capabilities. It's not like there's a fundamental trade-off between spelling Google correctly and understanding complex queries.

SPEAKER_01

This also raises questions about how these models are being evaluated and tested. Are companies so focused on benchmark performance that they're missing basic functionality issues?

SPEAKER_00

Oh, absolutely. The AI industry has this obsession with beating benchmarks and achieving high scores on standardized tests, but real-world reliability requires consistent performance on mundane tasks, too.

SPEAKER_01

And there's a user experience angle here too. If I'm using Google's AI for writing assistance, and it can't spell basic words, that completely undermines the value proposition. People expect AI to be better than humans at these kinds of mechanical tasks.

SPEAKER_00

Especially when you're competing with models like GPT-4 that generally handle spelling pretty well. This makes Google's AI look inferior on a very basic, visible dimension that everyone can understand.

SPEAKER_01

What's interesting is that this isn't even a new problem. Language models have had spelling issues for years, but somehow it feels more surprising when it's Google's AI doing it.

SPEAKER_00

That's because Google built its entire reputation on text processing and search. They literally organized the world's information. How can their AI not spell correctly? It violates our expectations about their core competency.

SPEAKER_01

Do you think this is fixable with current approaches, or does it point to fundamental limitations in how these models work?

SPEAKER_00

I think it's fixable, but it requires deliberate effort. You probably need specific training on spelling tasks, better tokenization strategies, or a post-processing to catch spelling errors, but Google apparently hasn't prioritized that work.

SPEAKER_01

There's also the question of whether this affects other languages. If the spelling issues are related to how the model handles character-level patterns, it might be even worse for languages that don't use the Latin alphabet.

SPEAKER_00

That's a great point. English spelling is already irregular and challenging for AI models. Languages with different writing systems or more complex orthographic rules could be even more problematic.

SPEAKER_01

What's the broader lesson here for people using AI tools in their work or personal life?

SPEAKER_00

Always fact-check and proofread AI output, even for simple tasks. Don't assume that because an AI can handle complex reasoning, it'll get the basics right. Trust but verify, especially for anything that's going to be public-facing.

SPEAKER_01

And for Google specifically, this is a reminder that they're still playing catch up in the AI race, despite their early research advantages. They need to get the fundamentals right before they can compete on advanced capabilities.

SPEAKER_00

It's also a reminder that AI development isn't just about raw compute power or cutting edge research. Sometimes the most basic functionality requires careful engineering and quality assurance.

SPEAKER_01

Let's move into rapid fire with some other big stories today. First up, former Google and Apple researchers have launched a startup called Trajectory that's focused on building AI's missing feedback loop through rapid iteration cycles.

SPEAKER_00

This is addressing a real problem. Most AI systems are trained once and then deployed, but they don't learn from real-world usage. Trajectory wants to create continuously learning AI products, which could be game-changing for enterprise applications.

SPEAKER_01

The fact that these are former Google and Apple researchers suggests they've seen the limitations of big tech AI development firsthand. When you have thousands of engineers and complex approval processes, innovation can slow down.

SPEAKER_00

Exactly. Like sometimes you need the flexibility of a startup to solve problems that big companies can't tackle due to bureaucracy or legacy systems. The rapid iteration approach makes sense when you're trying to build feedback loops that big companies might be too risk-averse to implement.

SPEAKER_01

What's interesting is the timing. Everyone's talking about AI agents and autonomous systems. But if those agents can't learn and improve from experience, they're always going to be limited by their initial training.

SPEAKER_00

Right, and this could be the missing piece that makes AI agents actually useful in complex business environments. Instead of static systems that break when they encounter unexpected situations, you get AI that adapts and improves over time.

SPEAKER_01

Next, Cisco and OpenAI are collaborating to use Codecs to redefine enterprise engineering. The partnership focuses on scaling AI native development, advancing AI defense initiatives, and automating defect remediation.

SPEAKER_00

This is huge because Cisco touches so much enterprise infrastructure. If they can successfully integrate codecs into their engineering workflows, it could dramatically improve the reliability and security of network systems.

SPEAKER_01

The AI defense angle is particularly interesting. Using AI to automatically detect and fix security vulnerabilities before they become problems. That's moving from reactive to proactive security management.

SPEAKER_00

Yeah. And given how much critical infrastructure runs on Cisco equipment, having AI that can proactively maintain and secure those systems could prevent major outages and breaches. This isn't just about productivity, it's about national security.

SPEAKER_01

The automated defect remediation part is fascinating too. Instead of human engineers having to track down bugs and write fixes, the AI could potentially identify issues and patch them automatically.

SPEAKER_00

That's the dream of self-healing systems. But it also raises questions about human oversight. How do you ensure that AI-generated fixes don't introduce new problems? The trust and verification challenges are massive.

SPEAKER_01

Still, for a company like Cisco that manages networking equipment in thousands of locations, having AI that can diagnose and fix problems remotely could save enormous amounts of time and money. OpenAI has pledged 250 million tolls to support workers affected by AI disruption. This initiative aims to help individuals adapt to rapid changes in the job market caused by artificial intelligence.

SPEAKER_00

This feels like OpenAI trying to get ahead of the inevitable backlash from job displacement. $250 million sounds like a lot, but spread across potentially millions of affected workers, it's probably not going to solve the fundamental problem.

SPEAKER_01

But it's at least an acknowledgement that AI companies have some responsibility for the societal impacts of their technology. That's a step forward from the usual innovation is always good rhetoric we hear from tech companies.

SPEAKER_00

True. And it sets a precedent for other AI companies to contribute to retraining and transition programs, though I'd rather see systemic policy solutions than just corporate charity.

SPEAKER_01

The interesting question is how they'll actually deploy this money. Are we talking about retraining programs, unemployment assistance, or funding for new types of education? The details matter a lot here.

SPEAKER_00

And there's a timing issue too. AI disruption is happening now. But retraining programs take years to show results. By the time someone completes a new certification, the job market might have changed again.

SPEAKER_01

It's also worth noting that this is coming from OpenAI specifically, the company that's probably most responsible for the current AI acceleration. There's definitely some self-interest in managing public perception here. Finally, early reports suggest Snowflake has committed $6 billion to Amazon Web Services to support global AI expansion. This investment underscores their strategy to leverage cloud infrastructure for AI-driven initiatives. It also highlights how expensive it is to compete in the AI space. These aren't software costs anymore. This is serious infrastructure investment. You need massive compute power to train and run modern AI models.

SPEAKER_00

Right. And it creates a competitive moat. Smaller companies that can't make billion-dollar infrastructure commitments are going to struggle to compete with the scale that Snowflake can now offer.

SPEAKER_01

There's also the strategic partnership angle here. By committing so much to AWS, Snowflake is essentially betting their future on Amazon's cloud infrastructure. That's a big dependency for such a large company.

SPEAKER_00

But it also gives them access to AWS's latest AI chips and services before competitors. When you're spending six billion dollars, you probably get some special treatment and early access to new technologies.

SPEAKER_01

This trend of massive infrastructure investments is reshaping the AI industry. It's not enough to have good algorithms anymore. You need the computational resources to deploy them at scale.

SPEAKER_00

And it's creating a new kind of vendor lock-in. Once you've invested billions in a specific cloud provider's infrastructure, switching becomes almost impossible. These partnerships are going to define competitive dynamics for years.

SPEAKER_01

If you zoom out and look at everything we covered today, there's a clear pattern emerging around AI becoming more autonomous and more integrated into critical systems, from trading stocks to managing enterprise infrastructure.

SPEAKER_00

Yeah. And and what's striking is how quickly we're moving from AI as a tool to AI as an agent. These systems aren't just helping humans make decisions anymore. They're making decisions independently and taking actions in the real world.

SPEAKER_01

But we're also seeing the growing pains. Google's spelling problems, YouTube's detection challenges, Meta's subscription experiments, these are signs that the technology is advancing faster than our ability to implement it reliably.

SPEAKER_00

The billion-dollar infrastructure investments from Snowflake and the Worker Displacement Fund from OpenAI suggests that companies know we're at an inflection point. They're preparing for a world where AI fundamentally changes how business operates.

SPEAKER_01

What's interesting is the tension between autonomy and accountability that runs through all these stories. Robin Hood's AI agents can trade stocks, but who's responsible when they lose money? Cisco's AI can fix network problems, but what happens when the fixes create new issues?

SPEAKER_00

And YouTube's automatic labeling system could help fight misinformation, but false positives could damage legitimate creators. There's this recurring theme of powerful capabilities coupled with uncertain consequences.

SPEAKER_01

The subscription models that Meta is rolling out also represent a broader shift in how AI gets monetized. Instead of just improving existing products, AI is becoming the basis for entirely new revenue streams.

SPEAKER_00

Right, and that creates new incentives for AI development. When AI features drive subscription revenue, companies have strong motivations to push capabilities forward quickly, potentially at the expense of safety or reliability.

SPEAKER_01

The international dimension is also becoming more important. Different countries are going to regulate AI differently. Companies like YouTube have to navigate varying disclosure requirements, and infrastructure investments like snowflakes will determine which regions get access to cutting-edge AI capabilities.

SPEAKER_00

And there's the human impact that OpenAI's worker fund acknowledges, but probably can't solve. We're automating cognitive work at an unprecedented scale, and the social systems for handling that transition just aren't ready.

SPEAKER_01

The question for 2026 is whether we can develop the regulatory frameworks, safety measures, and social support systems fast enough to keep pace with the technology. But also an exciting one. Despite all the challenges, the potential benefits of AI agents that can manage finances, create content, and maintain infrastructure are enormous. We just need to figure out how to realize those benefits responsibly.

SPEAKER_00

The key is going to be building systems that are powerful but also transparent, autonomous but accountable, and innovative but safe. That's a difficult balance, but getting it right could determine whether AI becomes a net positive for society. And if you're finding value in these daily deep dives into AI news, subscribe wherever you get your podcasts and share the show with someone who needs to stay informed about what's happening in AI.