There’s a growing problem we’re seriously underestimating: we’re teaching #AI to code, but we’re not teaching humans to code anymore. #SatyaNadella revealed that 30% of the code written at #Microsoft today is generated by AI, with an acceptance rate of 30–40%. Translation: nearly a third of the software from the most influential Big Tech company is now built by human-machine collaboration. Yet we’re getting used to it like it’s normal, without asking what happens if we forget how to build what we rely on every day.
Meanwhile, from China comes a reminder that size no longer matters. #Xiaomi launched an open-source model called #MiMo: just 7 billion parameters, yet performance comparable to much larger models like o1-mini on complex tasks like math and coding. It’s not about budget anymore. It’s about direction. And right now, China seems to have more of it than anyone else.
#JensenHuang, CEO of #NVIDIA, made it clear: China is “not behind” in AI. He named #Huawei as a key player catching up fast. But most importantly, he called it an “infinite race” — where you never truly win, you just stay in the game. It’s no longer about raw computing power; it’s about political will. And the West is already losing focus.
Speaking of power — real power — it’s worth watching what #MiraMurati is doing. The former CTO of #OpenAI now leads Thinking Machines Lab and is reportedly closing a $2 billion round. But the real news isn’t just the money. Some sources claim she’ll hold full control over board votes. In an era of omnipresent investors and collective strategies, that’s an act of defiance. And maybe the only real way to innovate: putting power back into the hands of builders.
Meanwhile, we’re starting to gamble away our identity, one pixel at a time. #Runway just launched #Gen4References: you can upload a photo, a selfie, or a 3D model and insert yourself seamlessly into any scene. Image by image, video by video, we’re becoming protagonists anywhere. It seems playful, but it’s a profound shift — because when your image can be cloned precisely by anyone, it also becomes easier to manipulate. And the line between representation and distortion grows thinner.
At the same time, #Freepik and #Fal made a decision that could have a huge impact: they released #FLite, a model trained only on licensed data. In a world rushing to build models without worrying about data sources, they chose to go clean. It’s harder, slower — but it might just make the difference when the real legal checks come.
And then there’s #Duolingo, moving at full speed. They just launched 148 new courses — the largest content drop in their history. But behind the scenes, the bigger news is that Duolingo now defines itself as an “AI-first” company. It’s no longer just a language learning app. It’s a massive content generation machine powered by AI. And that’s where the real battle lies: using AI not just to replace, but to multiply. As long as quality isn’t sacrificed — because learning badly is worse than not learning at all.
