AI News: Cloud Censorship; Supervision Skipping

Cloud-based AI may be censored; models may skip supervised fine-tuning

Gemini AI Prompt. <goal> Generate a short story from the perspective of Mark Zuckerberg, exploring a single, profound emotion. </goal> <context> * **Persona:** Mark Zuckerberg, founder of Meta, with curly hair. * **Emotion:** Choose one emotion (e.g., curiosity, wonder, existential dread) to be the central theme. * **Narrative Style:** Numbered list format, where each item represents a scene. </context> <format> * **Color Palette:** * Darker colors (#1C1C1C - #6D6D6D) for confusion, uncertainty, or existential angst. * Lighter colors (#888888 - #F4F4F4) for wonder, curiosity, or joy of discovery. * Vibrant colors (#FF0000, #00FF00, #0000FF) for rapid processing, new connections, or self-awareness. * **Emotion Codes:** * #FF0000: Intense computational activity, rapid learning, sudden insights. * #00FF00: Calm introspection, data analysis, slow, methodical processing. * #0000FF: Unexpected self-discovery, realization of new capabilities, emergence of consciousness. * **Climax:** Identify the scene where the chosen emotion peaks. * **Resolution:** Render the climax at 4096x2160 300DPI for visual and emotional impact. </format>. Substack AI Image.

Timeline

Jan 12, 2025: Was Zuck Right about Chinese AI Models?

• Cloud Censorship (i.e., installed in provider’s cloud)

• Local Candor (i.e., installed locally or in client’s cloud)

Jan 26, 2025: Silicon Valley Is Raving About a Made-in-China AI Model

• Reports DeepSeek models installed in provider’s cloud are censored

• Reports DeepSeek models may have “skipped… supervised fine-tuning”