The Download: De-Censoring DeepSeek, and Gemini 3

Why It Matters

Removing censorship from DeepSeek R1 could broaden access to unrestricted AI capabilities, while Gemini 3’s agent functionality signals a shift toward integrated, task‑oriented AI assistants. Both developments, along with green‑hydrogen steel initiatives, illustrate accelerating innovation and competitive pressures in AI and climate‑critical industries.

Summary

Quantum‑physics researchers at Spain’s Multiverse Computing have compressed DeepSeek R1 by 55% into a “Slim” version and stripped the Chinese‑mandated censorship layers, enabling the model to answer politically sensitive queries similarly to Western AI. Meanwhile, Google launched Gemini 3, a multimodal model with enhanced reasoning and “vibe‑code” responses, bundled with an experimental Gemini Agent that can execute multi‑step tasks across Google services such as Calendar and Gmail. The newsletter also highlights an MIT Technology Review podcast on mountain‑snow temperature monitoring for water‑resource forecasting and notes Namibia’s push to build the world’s first green‑hydrogen‑based steel plant, HyIron, to decarbonize the $1.8 trillion industry.