Chinese Voice Actors Forced to "Prove They're Human" — The Age of AI Voice Cloning
Chinese media outlet Sixth Tone has reported on a situation where voice actors must now prove they are human to compete against their own AI clones.
As AI voice cloning technology improves, cases of professional voice actors' voices being replicated without consent are on the rise. Beyond simply losing work, some actors now face the reverse suspicion: "your voice might be AI." How to protect voice ownership and identity is a serious challenge accompanying the spread of voice AI.
The problem extends beyond the voice acting industry. As AI-generated deepfake audio becomes commonplace, the challenge of proving a voice is "real" demands urgent solutions on both the legal and technological fronts.
Humans Cannot Identify AI-Generated Poetry — But Knowing It's "AI-Made" Kills the Enjoyment
A study published in the academic journal Digital Scholarship in the Humanities has revealed that humans cannot distinguish AI-written poems from human-written ones, while being told a poem was "written by AI" lowers their enjoyment of it.
Participants could not accurately identify which poems were AI-generated and which were written by humans, yet they uniformly rated poems labeled as "AI-made" lower. This shows that bias against AI-generated content operates independently of actual quality assessment.
The paradox that the better AI-generated content becomes, the relatively larger the impact of labeling grows, could also influence how transparency regulations for AI content take shape.
A "Sovereign" Open-Source Model for German and English
A paper titled "A Sovereign, Open-Source Foundation Model for German and English" has been released on Hugging Face. The initiative aims to build an open-source foundation model specialized in German and English, pursuing AI sovereignty in Europe.
In Europe, alongside the enforcement of the AI Act, calls are growing to reduce dependence on US and Chinese AI models. Developing AI models grounded in one's own data and values is increasingly regarded as essential from a technological sovereignty perspective.
DeepSeek v4 Flash vs Mimo 2.5 — Comparing New Local LLMs
Over on Reddit's LocalLLaMA community, a comparison between DeepSeek v4 Flash (text-only) and Mimo 2.5 (omnimodal) is under discussion.
DeepSeek v4 Flash is fast and specialized for text processing, while Mimo 2.5's strength is its multimodal support. According to user impressions, DeepSeek Flash feels faster, but Mimo 2.5 earns points for being omnimodal.
The discussion confirms once again that the right model depends on the use case — a sign that the range of options for local LLMs keeps widening.
Baton: A Management Tool for AI Coding Agents
A tool called "Baton" for managing AI coding agents has been released on GitHub. When running multiple AI coding agents in parallel, it helps you keep track of which agents require human intervention.
As AI agents grow more autonomous, designing effective "human-in-the-loop" workflows — pinpointing the right moment for a person to step in — is a critical challenge. Baton visualizes agent status and aims to direct human attention to the tasks that need it most.
AVA: A Self-Hosted Voice AI Agent
A self-hosted voice AI agent for Asterisk/FreePBX, called "AVA," has been released on GitHub. By integrating an AI voice agent into an open-source PBX system, you can build a voice response system that does not depend on external APIs.
Driven by privacy and cost considerations, demand for self-hosting voice AI is on the rise. Efforts like this, bridging traditional telephony systems and AI, offer companies a new option for automating customer support.
---