🚀 The Current State of AI 🤖 Matías S. Zavia 🔥 🤖 “AI took my job, but gave me back my passion”

🎙️ “ChatGPT knows my innermost secrets (and so does my wife)” 🧠 “I don’t think AI is conscious, but it’s far too useful to ignore”

Matías S. Zavia (Concordia, Argentina, 1989) is a telecommunications engineer; from Málaga he contributes to Xataka and co-hosts the AI podcast Monos Estocásticos with Antonio Ortiz, winners of Parèntesi MÈDia’s 2024 GEN-AI Awards. He worked at Weblogs and Gizmodo España… until an AI took his job. He’s one of the most approachable, expert and comprehensible voices on AI in Spain. You can read and watch the full interview in Transparent Algorithm.

— You’re in Barcelona, without Antonio, to receive the award.
— It feels strange not to have him here, yes. He was my boss, now we’re friends, and there’s an unresolved sexual tension between us. We do CrossFit together and there are very intimate exercises, like the bench press.

— Does sweating and that musk of masculinity bring you closer?
— He gets the worst part, because I sweat more than he does and I often forget my towel at home.

— Don’t you have an AI to remind you?
— You have no idea how much of a blessing AI is for chaotic, disorganized people like me. You can have all your files in a mess, but AI keeps getting better at finding what you’re looking for. It helps me, for sure.

— Doesn’t AI stress you out with so many new tools?
— There comes a point when it’s overwhelming, Saül, because so many things come out every week that not only don’t fit in our podcast script, but sometimes at night I think, “I should be testing the latest stuff instead of watching a movie with my wife.” FOMO (fear of missing out) gets easier to manage with maturity, little by little.

— At the same time, don’t you think that if we don’t dive in now, we’ll regret it later?
— The energy transition is already proving to be heavier than the Industrial Revolution. We’ve always seen that as the world-changing event, but we’re living a revolution that could be even more powerful. I’ve evolved because AI has gone through several winters: five years ago I was a skeptic. When everyone talked about AI, I anthropomorphized it, saying “watch out, there’ll be a machine smarter than humans.” But now I realize that reaching human-level intelligence wasn’t so far off—if AIs can compose poems that move you, perhaps creativity wasn’t something exclusively human. A language model can produce a poem that truly touches you. I’m all aboard this train now.

— And what about AI-generated music?
— That’s a problem. I’m so hooked on AI-made songs that my wife has banned me from using it. She’s had enough. The other day Spotify blocked me because I was creating too many AI tracks. My distributor said they didn’t want to work with me anymore. When someone with no talent but creative drive like me sees Suno or Udio arriving, I can’t sleep: I spend all night generating songs, then I want to share them on Apple Music and Spotify. It feels like some record label has blacklisted me.

— Is AI causing you marital problems? Have you felt a speed gap with AI?
— I always say that if I die, please let nobody find my Google history or my ChatGPT history, because in ChatGPT there are intimate conversations about medical issues or worries you’d be embarrassed to share. I gave my ChatGPT password to my wife—it’s like handing over my whole world—because she was skeptical about AI, and now she’s incorporated it into her work as a teacher. There are tons of tools she can use with her students for activities. We use ChatGPT every day, each on our own project. I’d say AI has actually brought us closer.

— Conversational consumption is more intimate than Google—aren’t you afraid she has your ChatGPT password?
— Yes, but consider this: before I had advanced voice mode, I used chatbots very utilitarian-style: for work, to summarize documents or correct articles. Since I set up an iPhone shortcut to talk to ChatGPT, I press the button, ChatGPT’s voice plays, and the queries are totally different. That phenomenon of using AI as a therapist or for cooking advice has entered a delicate territory I always found a bit odd.

— You do it out of conviction, not just because you need it for the podcast, right?
— I do it because I get so much out of it. I’m a bit scattered, and where I can’t reach, AI helps me get there.

— Toni Esteve, from Parèntesi MÈDia—who’s giving you the award—defends the concept of “augmented human intelligence.”
— Many experts reject the term “artificial intelligence.” We still have so much to understand about the human brain and how consciousness arises that it feels wrong to attribute it to stones, silicon, chips… that black-box of probabilities giving us answers. I’m skeptical that AI will ever become conscious or human-like. I use it as augmented intelligence, as Toni says, and if I worry about security it’s not about a Matrix-style scenario, but that machines get so good that the companies holding the best AI become too powerful, instead of having AI open and available to everyone. That’s my sole big concern.

— Is there any company you’re particularly wary of for the future?
— OpenAI operated almost like a sect: they only hired people who believed in AGI (Artificial General Intelligence) capable of matching humans. But they’re hitting their milestones: they promised AI agents by 2025, and we’re already seeing them—AIs that “act” for you, not just respond to your instructions. Now they say they have an AI innovator for science. We’ll get to human level, though with today’s transformer/LLM paradigm there’s a wall: hallucinations. They make mistakes frequently and propagate them through the conversation. I don’t see OpenAI concentrating all the power; rather, Google, DeepSeek or Anthropic push them so hard they release models hastily. They’re not the big evil-corp from a superhero movie. I’m glad competitors like Meta or DeepSeek embrace open, customizable AI.

— What do you think of Ilya Sutskever and Mira Murati after their “coup” at OpenAI?
— When they ousted Sam Altman, it seemed like fear of what they had built and doubts about his ability to steer it without wrecking civilization. But their most powerful model, o3, isn’t dangerous: it still makes errors and can’t replace a senior programmer. I don’t understand that coup, nor how a company like Sutskever’s is so well funded without a clear product. Murati, on the other hand, has a more user- and business-oriented approach, which I get better. Ilya is the scientific genius, but Murati inspires more confidence for the commercial side.

— Is it really only Sam Altman and Greg Brockman who can carry the carriage?
— To retain talent, even Google pays a full year’s salary to someone taking sabbatical so they don’t join competitors or strike out on their own. More than energy, the issue is talent: in San Francisco, everyone working in AI is a multimillionaire. Leadership doesn’t seem to be the problem.

— Sam Altman is a marketing genius…
— I’m not sure I love that “guerrilla marketing,” but he’s great at it. Sometimes you see the seams, but it works.

— Does he worry about Elon Musk breathing down his neck?
— They have an open feud. In the end, what matters is who has the best model and best value proposition. Marketing can buy you a few months, but the most powerful model wins. When I saw Gemini 2.5 Pro spoke Spanish better than ChatGPT, I switched instantly. No one’s tied to ChatGPT.

— And what about Musk’s AI, Grok, on X?
— I’m impressed how quickly Grok has leveled up: in one or two years it’s on par with the best. If you’ve got the richest man in the world building massive server farms and recruiting talent, you can catch up fast. I use Grok a lot—I’m a Twitter [X] addict—and the integration is excellent. Being able to pull real-time info from Twitter is its big edge. Maybe that’s why rumors say OpenAI wants its own social network.

— OpenAI’s search is a leap forward, but it lacks social-media immediacy.
— Sometimes ChatGPT’s search feature kicks in and I get annoyed because I know the answer won’t be as good as its trained model. A chatbot needs up-to-the-minute data. Search is mandatory, and how it’s integrated hugely affects quality. ChatGPT uses Bing, so results vary by source. My biggest worry is that real-time info comes from journalists like us—and we won’t see a cent because AI ingests and explains content without the user clicking on the original outlet or ads. There must be a licensing system for journalistic content, or it’s unsustainable.

— How’s your experience with Claude, Anthropic’s AI?
— Claude just got updated with reasoning capabilities, surpassing earlier paradigms. Even its base models were favorites among developers: its “one-shot” code generation—write a prompt, get flawless code in one go—blows programmers away. For my long-form writing, though, it feels overly flattering. This shows that the difference isn’t just in pre-training—fine-tuning matters. Anthropic nails fine-tuning for programming. The AI race outcome remains wide open: Claude led a few months ago; now OpenAI sits atop the podium.

— Meta claimed Llama 4 was best, then disappointed. How do misleading AI announcements affect users?
— Meta’s stunt is classic: they hyped Llama 4, then it underdelivered. They already had a solid Llama 3—no need to rush. But the race is so frantic and cash-fueled that everyone pretends to be first. This fuels premature launches: OpenAI’s o3 was also rushed. Google even gives away Veo 2, the top video generator, so people use their tools. Desperation breeds freebies.

— How do you assess the impact of a possible Trump return to the White House on AI? Has Europe—and its tariffs—woken up?
— Exactly. We knew China could catch up with its massive state resources; DeepSeek was a shockwave: if you didn’t try it, you missed out. My dad texted me, “Download it—it’s free and rivals the best!” I’m not sure if DeepSeek or the ensuing trade war was bigger, but I’m thrilled Europe answered with public-private partnerships to build open, sovereign, language-tailored AI. An AI trained on the entire Catalan literary canon could be the world’s best in Catalan. That’s how we achieve technological sovereignty.

— Is Mistral AI, France-backed start-up, the model to follow?
— Unlike proprietary models, open ones let you catch up faster via distillation and fine-tuning. Meta’s Llama has billions of parameters. If you adapt open models in Europe with those techniques, any company or individual can deploy a powerful model with moderate investment. Mistral, as a private start-up, could build Europe’s top models, which later benefit from private R&D if licenses allow. I believe in both paths: a strong private player like Mistral and public supercomputers in Spain for free use.

— How do you see Spain in the geo-strategic AI landscape?
— Spain and Europe aren’t on par with the US or China, but that’s not the goal. Europe won’t birth Silicon Valley-scale giants, but it can have its own projects and models tailored to local needs. Even if smaller, they’ll be ours, and we’ll gain sovereignty.

— Thanks to Trump’s tariffs, has Europe sidelined ethical debate for competitiveness?
— There’s been a moral swing: things we took for granted aren’t so clear anymore. Europe is swinging, but keeps a middle ground. With Trump we feared a “Milei 2.0” US tilt, yet Europe’s in balance. Sure, Americans mock our plastic bans, but the US is reducing obesity thanks to Ozempic, a European drug. Europe has huge talent and industry. I think EU AI regulation won’t be as strict as once proposed, but we’ll remain leaders in renewables, climate action, space debris cleanup, universal chargers and removable batteries. A cycle shift toward some deregulation is coming.

— Monos Estocásticos turns two years old, right?
— Yes, we launched shortly after ChatGPT’s debut in November 2022. The toughest part was the name: we tested it with family and no one liked it except my sister. In the end it stuck—though some still call us Monos Escolásticos.

— Two years on, how would you gauge public interest in AI?
— In Spain skepticism runs high because negative news sells better than positive. I know more and more journalists use AI and find it useful; I attribute the disconnect to the state of journalism and the priority of eye-catching headlines over journalists’ real opinions.

— Does that not reflect a genuine fear among some pros?
— They have good reason: low-quality journalism—which sadly dominates Google—is content that doesn’t need trained, passionate writers. There are probably more fans of Manuel Jabois than ChatGPT, though the latter is catching up. I believe Jabois will keep his job.

— How do you envision a Sant Jordi in ten years dominated by AGI?
— Sant Jordi across all Spain left a mark on me. I didn’t use AI even once that day… except to make sure I didn’t overpay for my rose: I bought it for €6 while the stall next door sold them for €3.50.