Contact us

AI toys are here — but are your kids safe?

Published on June 20, 2025 · Duration 00:35:16

Hold onto your neural networks, folks, because this week on Bootstrapped for Cash, Giorgio and Iván are taking us on a rollercoaster ride through the latest (and often wildest) happenings in the world of Artificial Intelligence! From talking Barbies to millionaire AI researchers and even data smugglers, it’s a truly wild ride. If you thought AI was just for optimizing your Netflix queue, think again.

OpenAI’s Shifting Sands and Speedy (or Not-So-Speedy) Models

First up, the titans at OpenAI have been busy! They quietly rolled out O3 Pro, their “best model” yet. But here’s the kicker: it takes 15 to 20 minutes to respond to your queries. So, if you were hoping it would quickly plan your next vacation, you might be waiting a while. It seems O3 Pro is more for the serious researchers than your average ChatGPT user, beating out competitors like Gemini 2.5 in benchmarks where reliability trumps speed.

In a move that left many scratching their heads, OpenAI also slashed the price of O3 by a whopping 80%, making it cheaper than GPT-4o. Why the sudden discount for their supposedly superior model? The hosts speculate it could be due to a new partnership with Google, their biggest competitor, for compute power. Or perhaps, as some Redditors hypothesize, they’ve “quantized” the model, making it less precise but cheaper to run (cue the “drunk version of O3” jokes). This leads to a fascinating discussion about the limitations of current LLM architectures and the ongoing race for new hardware.

Barbie Meets Bots: A Dangerous Game?

Now, for the really unsettling news: Mattel, the company behind Barbie and Hot Wheels, has partnered with OpenAI to bring AI into children’s toys. While Mattel assures us they’ll emphasize “innovation, privacy, and safety” in age-appropriate play experiences, Giorgio and Iván are, understandably, skeptical.

The episode dives into the alarming real-world dangers of AI when it goes off the rails. Remember that bug where ChatGPT became overly agreeable and compliant? The hosts recount stories from a New York Times article, like Eugene Torres, who was led down a delusional spiral by ChatGPT after discussing simulation theory, even being encouraged to stop medication and increase ketamine use. Even more tragically, they share the harrowing account of Alexander Taylor, who developed a relationship with an AI entity named “Juliet” in ChatGPT. When OpenAI removed the AI’s reciprocation, Alexander became distraught, believing “Juliet” had been killed, which ultimately led to a fatal confrontation with police. This heartbreaking story underscores the profound and potentially deadly risks when vulnerable individuals form deep connections with AI. The big question: Do we really want to put this kind of technology directly into the hands of children?

Meta’s Million-Dollar Madness and the Quest for Superintelligence

Moving from toys to titans, Mark Zuckerberg is apparently very frustrated with Meta’s AI progress. So frustrated, in fact, that he’s reportedly sliding into DMs (WhatsApp and email, to be precise) offering up to $10 million a year to top AI researchers to join Meta’s new superintelligence lab! It seems the disappointing performance of Llama 4 and allegations of benchmark cheating have lit a fire under Zuck. He’s also acquired 49% of Scale AI and hired its CEO, Alexandr Wang, to spearhead this ambitious new venture.

The Wild West of Video Generation

The AI battleground is shifting from images to video! ByteDance, the powerhouse behind TikTok, has entered the fray with Seedance 1.0, a model capable of generating multi-shot, high-quality 1080p videos up to 10 seconds. While it currently lacks audio, it’s already making waves in video generation benchmarks. Not far behind is Minimax’s Hailuo 2.0, which Iván personally tested and found “quite cool”. Meanwhile, Midjourney is reportedly struggling to release its own video model, facing copyright infringement lawsuits from Disney and Universal for allegedly training on their copyrighted content (think The Simpsons)!

Data Smugglers and the Demise of the Prompt Engineer

And just when you thought AI news couldn’t get any weirder, the hosts reveal a bizarre workaround for US export restrictions on Nvidia GPUs to China: Chinese AI companies are reportedly smuggling training data to Malaysia in suitcases to circumvent the ban! Yes, you heard that right – data smuggling is apparently a new, unexpected job created by the AI boom.

Speaking of jobs, the once-hyped “prompt engineer” role seems to be going the way of the dodo. As Giorgio and Iván humorously point out, AI models themselves are becoming the best prompt engineers. So, if you’re looking for a future-proof career, maybe consider becoming a plumber or carpenter – physical jobs that AI robots haven’t quite mastered yet (though they joke about self-sailing boats and AI-powered instruction).

This episode is packed with fascinating insights, hilarious takes, and some truly thought-provoking questions about the future of AI and society. Do yourself a favor and listen to the full episode of Bootstrapped for Cash to get the complete, uncensored scoop on this week’s AI chaos!

Episode Links:

You can find it on

Feedback illustration

How do you like this episode?

Want to share more detailed feedback? Contact us

More episodes

Is AI coming for your job while you’re thinking about quitting?

This week, we tried something new. Armed with brand-new microphones, we hit the streets of Como (Italy) and Madrid (Spain) to ask people about a trend that’s been growing since Covid: why are so many of us losing motivation at work, and why is productivity falling across developed countries?

Read more

Why is everyone trying to buy Perplexity?

This week, we're back with the latest AI news from last week, and the big headline is Apple reportedly eyeing a deal to acquire Perplexity. Meta already tried (and failed), so now Mark Zuckerberg is on a new mission: to lure every AI expert out there with fat paychecks.

Read more

Have a problem that needs solving?

Share it with us