Block's Dessa Full-Duplex Voice AI Shocks Industry; Ex-DeepMind Researcher Puts 10-Year Human Extinction Risk at 33% - AI Daily Brief
Audio in Mandarin Chinese · English transcript below
⚡ Block's Dessa enables full-duplex voice, Kimi K3.1 Pro advances open Agent, MiniMax & DeepSeek cleared for global expansion—ex-DeepMind researcher puts AI extinction at 33% this decade. Breakthroughs and risks hit max simultaneously.
AI is undergoing a dual acceleration toward openness and anthropomorphism. Kimi's push for open Agents and Block's achievement of full-duplex dialogue through pure voice training mark an approach to the natural inflection point in human-computer interaction. Yet the warning from a former DeepMind researcher—placing a one-in-three probability on human extinction within a decade—casts the safety risks behind this technological surge into stark relief.
Today's Top 3 Headlines
- AI Industry News
🤖 Block's Dessa: Full-Duplex Voice Model, Pure Speech Training, Single GB200 System, Runs on M4 Max Locally
Block's lab drops Dessa: pure speech-to-speech training, single GB200 system, runs locally on M4 Max in real-time. End-to-end voice training path is now viable; consumer-grade on-device hyper-realistic AI voice agents are here.
Source ↗ - AI
🤖 Ex-DeepMind Researcher: AI Has 1-in-3 Chance to Destroy Humanity Within Decade
Former DeepMind researcher revealed at World Summit: 33% chance AI destroys humanity within decade. For AI developers and investors, safety alignment investment must accelerate in lockstep with capability races, as extreme risk prevention shifts from fringe to core agenda.
Source ↗ - AI Industry News
🤖 Kimi Drops K3.1 Pro Tech Blog, Doubles Down on Open Intelligence
Kimi's K3.1 Pro technical blog proposes an "Open Intelligence" route, sparking heated Reddit discussion. For developers, Chinese LLMs now offer a powerful new open-source option; for OpenAI, its closed-source moat faces a collective challenge from community-driven open intelligence.
Source ↗
+6 more headlines
- 🤖 MiniMax & DeepSeek Approved: Open-Source LLMs Complete Export Registration for Global Use
- 🤖 llama.cpp New PR: qwen4exp QSA Selection Without kq_mask
- 🤖 Nvidia drops Nemotron-3.5-Bolt: 8B params in 4GB, targets edge local Agent
- 🤖 NHTSA Proposes FMVSS 108 Update: Cyan Lights for Robotaxis
- 🤖 MIT's Ramesh Raskar Challenges AGI Assumptions, Proposes Agent Collaboration Paradigm
- 🤖 Google updates privacy policy, adjusts limits on Gmail data for ad personalization
Podcast highlight
Alright, let's wrap this JustSayAI AI Daily Brief right here. Take your weekend slow.