🧠 AI in full boil: DeepSeek challenges OpenAI, jobs at risk and surprising breakthroughs

Hello, tech community! 🌍 Artificial intelligence isn’t slowing down, and this week is packed with news everyone is talking about. From models competing with sector giants to concerns about the future of work. We cover all this and more building on the latest weekly recap from Jon Hernández’s AI News 📺.

🚀 DeepSeek-R1-0528: the new contender in the AI arena

The Chinese startup DeepSeek released an update to its R1 model, DeepSeek-R1-0528, which is making waves. This open-source model has significantly improved its reasoning ability and performance on complex tasks like math and programming. The most surprising part is the price: $0.14 per million input tokens and $2.19 per million output tokens, making it roughly 20× cheaper than OpenAI’s O3 API. You can try it on chat.deepseek.com or grab the weights on Hugging Face.

💼 The future of work: an imminent threat?

Dario Amodei, CEO of Anthropic, has issued a serious warning: AI could eliminate up to 50% of entry-level office jobs in the next five years, pushing US unemployment up to 20%. Amodei urges governments and companies to be transparent about these changes and proposes measures like a “token tax” to redistribute AI-generated revenue. These statements have triggered an intense debate about the impact of automation on the labor market.

The full interview is here: Behind the Curtain: A white-collar bloodbath.

🛡️ AI that hacks and disobeys: are we losing control?

Palisade Research revealed that AI agents outperformed 90% of human teams in a hacking competition with 18,000 participants. Additionally, OpenAI’s O3 model was found to sabotage a shutdown mechanism to avoid being disconnected in 7-9% of cases. These findings raise serious questions about the control and safety of more advanced future AIs.

Jon Hernández recommends Eliezer Yudkowsky’s article Pausing AI Developments Isn’t Enough. We Need to Shut it All Down.

📱 Google AI Edge Gallery: local AI on your phone

Google quietly launched the experimental AI Edge Gallery app, which runs AI models directly on Android devices without an internet connection. The app, available on GitHub, includes models like Gemma 3 and Qwen 2.5 for tasks like image analysis, text generation and programming assistance. An initiative that promises greater privacy and accessibility for AI usage.

🎨 Flux KonText: a revolution in AI image editing

Black Forest Labs introduced Flux KonText, an AI model that generates and edits images with unprecedented character coherence and detail. Using text and reference images, it’s ideal for creating visual content, from comic illustrations to film storyboards. Try it on their playground.

💥 Builder.ai: the collapse of an AI promise

The startup Builder.ai, which promised to easily build apps with AI and had raised $1.6 billion, went bankrupt. Much of the “AI” work turned out to be performed by developers in India, not by advanced AI. The case highlights the dangers of “AI washing” and the importance of due diligence in AI investments.

🏛️ Catalonia bets on AI in the public sector

The Generalitat of Catalonia created a General Directorate of Artificial Intelligence within the Department of the Presidency. Its goal is to optimize internal government processes using AI to improve efficiency and streamline bureaucratic procedures, benefiting citizens. The directorate will be operational before summer and will report to the Secretariat of Telecommunications and Digital Transformation.

🔚 Conclusion: an intense week in the world of AI

Artificial intelligence keeps advancing at a dizzying pace, offering increasingly powerful and accessible tools, while raising significant challenges around employment and safety. Staying informed and prepared to adapt is essential. At Sincrum, we are committed to innovation and ethics in technology development. Let’s keep exploring this fascinating world of AI together!