AI 日报 — 2026-08-09

DeepSeek-V4 Latent Reasoning introduces a self-contained model, DeepSeek-V4-Flash-0731-Latent-Reasoning, designed to perform "thinking" within its latent space. This model is NVFP4-quantized and provided in a production vllm form, suitable for serving runtime. It represents an advancement in moving complex reasoning processes into the latent space, as highlighted in its Show HN presentation.

Anthropic is setting Claude Code's auto mode as the default, citing its superior safety performance. In testing, auto mode identified 89% of harmful actions, significantly outperforming manual review, which caught only 13.6%. This disparity is partly attributed to users approving 97% of permission prompts during manual review, suggesting a habitual response. This change highlights a move towards automated safety measures in AI development.

An OpenAI strategist suggests that AI labs should rival government power, highlighting a long-standing debate where technology companies develop powerful systems while governments struggle with regulation. The core issue is how much oversight AI can tolerate without hindering innovation or competitiveness. Unlike governments, AI companies lack democratic accountability, making decisions on leadership, terms of service, and product access without public due process. This raises questions about the future of public services if economic activity shifts towards privately controlled machine intelligence.