A "GPT-3 Moment" for Voice AI? The Truth About Dessa

Block just announced Dessa, claiming it is the most authentic full-duplex voice model ever built. Its lab lead went even further, calling it a "GPT-3 moment for speech." Even wilder: Block claims Dessa was trained entirely on raw conversational audio with zero text tokens and no pretrained language models. Whenever you tell a traditional voice assistant to "wait" mid-sentence, it either talks over you or freezes in awkward silence. Is Dessa a real architectural leap for conversational AI, or just a polished demo? In this breakdown, we examine how full-duplex speech models hear and speak simultaneously, why Meta's 2022 textless voice model failed, and what Block still hasn't proven. 🔗 Sources & Documentation: • Block Official Dessa Announcement: https://block.xyz/dessa • Ragavan Thurairatnam Announcement Post: https://x.com/RagavanThurair/status/2113671324400861645 • Kyutai Moshi Full-Duplex Speech Paper (arXiv 2410.00037): https://arxiv.org/abs/2410.00037 • Meta & INRIA dGSLM Paper (arXiv 2203.16502): https://ar5iv.labs.arxiv.org/html/2203.16502 • Jack Dorsey Announcement Post: https://x.com/jack/status/2113697902916751661 • Crypto Briefing Coverage & Analysis: https://cryptobriefing.com/block-unveils-dessa-ai-voice-model/ 🔔 Subscribe: https://www.youtube.com/channel/UC0DZj1PNa_Fp0MU6uPSKv5w?sub_confirmation=1 💙 Become a Member: https://www.youtube.com/channel/UC0DZj1PNa_Fp0MU6uPSKv5w/join 🐦 Twitter/X: https://x.com/cloud_codes 💬 Discord: https://discord.gg/mhgXXmhrh #ai #voiceai #dessa #block #machinelearning #speechtospeech #artificialintelligence #cloudcodes