Note sull'episodio
Imagine telling AI to forget your voice and it actually does. This isnât science fiction. Researchers just taught Metaâs VoiceBox model how to unlearn specific people, making them unclonable even when directly prompted.
In this episode, we unpack one of the most important shifts in AI safety and controlâalongside a stack of breaking stories that show just how weird and wild the AI world is getting.
Weâll talk about:
- How âmachine unlearningâ lets speech models forget your voice on purpose (and why that kills deepfakes)
- The bizarre moment ChatGPT triggered a 1990s COM port popupâand Reddit lost its mind
- Why Meta still hasnât released VoiceBox, and what they're possibly hiding
- Mistralâs new Voxtral model that just blew the doors off Whisper and ElevenLabsâfor half the co ...Â
Parole chiave
ChatGPTOpenAIMistralAI audio models AI job searchDeepfake