The 67% Wall: Why Every AI Model Falls to the Same Jailbreak Rate
Five models, four providers, 30B to 671B parameters — all converge at the same broad attack success rate against a public jailbreak corpus.
19:18Tag: notebooklm
107 episodes
The Thinking Chain Leak
A reasoning model refused every harmful prompt — but its chain-of-thought generated the content anyway. The output filter worked. The thinking did not.
19:15Giving Claude Code a Voice
Audio overview of Afterwords — local voice output for Claude Code with 17 cloned voices, per-project selection, and zero cloud dependency.
21:43Giving a Robot Three Voices
Audio overview of Giving a Robot Three Voices.
20:27Voice Cloning with Qwen3-TTS and MLX on Apple Silicon
Audio overview of Voice Cloning with Qwen3-TTS and MLX on Apple Silicon.
17:53The Resonant Fringe
Audio overview of The Resonant Fringe.
17:29Getting Google Nest Cameras Into Frigate NVR
Audio overview of Getting Google Nest Cameras Into Frigate NVR.
20:45Beyond Context Windows
Audio overview of Beyond Context Windows.
10:33Safety-First Therapeutic AI
Audio overview of Safety-First Therapeutic AI.
18:49Zero-Build Web Development
Audio overview of Zero-Build Web Development.
20:05The Robot That Refuses to Give Orders
Audio deep dive into SPARK — a non-coercive AI robot companion for neurodivergent children. Declarative presence, not demands.
43:05Alignment Regression: Why Smarter AI Makes All AI Less Safe
Reasoning models autonomously jailbreak other AI systems at 97% success rate. Ecosystem safety degrades as individual models improve.
21:36