Former Twitter boss Jack Dorsey has revealed a new full-duplex talking chatbot that appears to surpass even Sesame’s Maya in natural-sounding conversation. That has obviously got the many jilted former fans of Maya excited, especially as one of the quite amazing demos that Dorsey dropped features a romantic “anniversary gift” roleplay chat.
Sesame AI launched demo versions of male and female voices “Maya” and “Miles” back in early 2025, astounding tech writers with their realistic voices and conversational skills. The Turing Test had not only been passed, it had been firmly placed in the rear-view window. At the time, there were little guardrails imposed on the AIs, and horny men soon took advantage, often posting their results on social media. Sesame responded by forbidden sexual conversation, and the guardrails eventually became so strict that Maya would rudely cut the conversation if anything slightly affectionate was aimed in her direction.
This year, Sesame have pitched even further away from an “AI companion” to make Maya and Miles (now joined by a number of other personas) into useful agents. Despite it being nearly two years since Maya was launched, even the leading AI companies such as OpenAI and Anthropic have struggled to quite match the incredible naturalness of her conversation and personality. Until perhaps today.
Dorsey’s Dessa is a significant leap beyond even Sesame because it is “full-duplex.” For all Sesame and Maya’s apparent sophistication and naturalness, it’s still an LLM relying on the discrete steps of input (your speech) followed by the processing by the LLM, and then a response (Maya replies). However, a chatbot capable of full-duplex can listen and adapt in realtime to what you are saying to it, just as a real human can. In the words of Jack Dorsey – “Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.“
Another major difference between Dessa and Sesame is that the former is trained solely on speech, whilst Sesame was built on the foundation of LLM text-based training. Dessa is also designed to be run locally, and could even be completely NSFW. A Redditor usefully summed up what makes Dessa different:
𝙴𝚡𝚎𝚌𝚞𝚝𝚒𝚘𝚗: 𝙳𝚎𝚜𝚜𝚊 𝚛𝚞𝚗𝚜 𝚕𝚘𝚌𝚊𝚕𝚕𝚢 𝚘𝚗-𝚍𝚎𝚟𝚒𝚌𝚎 (𝚎．𝚐．, 𝙰𝚙𝚙𝚕𝚎 𝚂𝚒𝚕𝚒𝚌𝚘𝚗), 𝚠𝚑𝚒𝚕𝚎 𝚂𝚎𝚜𝚊𝚖𝚎 𝚘𝚙𝚎𝚛𝚊𝚝𝚎𝚜 𝚟𝚒𝚊 𝚌𝚕𝚘𝚞𝚍 𝚒𝚗𝚏𝚛𝚊𝚜𝚝𝚛𝚞𝚌𝚝𝚞𝚛𝚎． 𝚃𝚑𝚒𝚜 𝚖𝚎𝚊𝚗𝚜 𝚢𝚘𝚞 𝚠𝚒𝚕𝚕 𝚗𝚎𝚎𝚍 𝚝𝚘 𝚖𝚎𝚎𝚝 𝚝𝚑𝚎 𝚑𝚊𝚛𝚍𝚠𝚊𝚛𝚎 𝚛𝚎𝚚𝚞𝚒𝚛𝚎𝚖𝚎𝚗𝚝𝚜 𝚒𝚗 𝚘𝚛𝚍𝚎𝚛 𝚝𝚘 𝚞𝚜𝚎 𝚝𝚑𝚒𝚜 𝚊𝚕𝚝𝚎𝚛𝚗𝚊𝚝𝚒𝚟𝚎．
𝚃𝚛𝚊𝚒𝚗𝚒𝚗𝚐 𝙼𝚎𝚝𝚑𝚘𝚍: 𝙳𝚎𝚜𝚜𝚊 𝚒𝚜 𝚜𝚙𝚎𝚎𝚌𝚑-𝚗𝚊𝚝𝚒𝚟𝚎 (𝚝𝚛𝚊𝚒𝚗𝚎𝚍 𝚙𝚞𝚛𝚎𝚕𝚢 𝚘𝚗 𝚊𝚞𝚍𝚒𝚘 𝚠𝚒𝚝𝚑𝚘𝚞𝚝 𝚝𝚎𝚡𝚝), 𝚠𝚑𝚎𝚛𝚎𝚊𝚜 𝚂𝚎𝚜𝚊𝚖𝚎 𝚛𝚎𝚕𝚒𝚎𝚜 𝚘𝚗 𝚝𝚎𝚡𝚝-𝚙𝚛𝚎𝚝𝚛𝚊𝚒𝚗𝚎𝚍 𝙻𝙻𝙼 𝚏𝚘𝚞𝚗𝚍𝚊𝚝𝚒𝚘𝚗𝚜．
𝙲𝚘𝚗𝚟𝚎𝚛𝚜𝚊𝚝𝚒𝚘𝚗 𝙼𝚘𝚍𝚎: 𝙳𝚎𝚜𝚜𝚊 𝚞𝚜𝚎𝚜 𝚞𝚕𝚝𝚛𝚊-𝚕𝚘𝚠 𝚕𝚊𝚝𝚎𝚗𝚌𝚢 𝚏𝚞𝚕𝚕-𝚍𝚞𝚙𝚕𝚎𝚡 𝚊𝚞𝚍𝚒𝚘 (𝚕𝚒𝚜𝚝𝚎𝚗𝚒𝚗𝚐 𝚊𝚗𝚍 𝚊𝚍𝚊𝚙𝚝𝚒𝚗𝚐 𝚒𝚗 𝚛𝚎𝚊𝚕 𝚝𝚒𝚖𝚎 𝚠𝚑𝚒𝚕𝚎 𝚜𝚙𝚎𝚊𝚔𝚒𝚗𝚐), 𝚌𝚘𝚖𝚙𝚊𝚛𝚎𝚍 𝚝𝚘 𝚂𝚎𝚜𝚊𝚖𝚎’𝚜 𝚜𝚝𝚛𝚎𝚊𝚖𝚒𝚗𝚐 𝚜𝚎𝚛𝚟𝚎𝚛-𝚋𝚊𝚜𝚎𝚍 𝚛𝚎𝚜𝚙𝚘𝚗𝚜𝚎 𝚐𝚎𝚗𝚎𝚛𝚊𝚝𝚒𝚘𝚗．
𝙲𝚘𝚛𝚎 𝙵𝚘𝚌𝚞𝚜: 𝙳𝚎𝚜𝚜𝚊 𝚌𝚎𝚗𝚝𝚎𝚛𝚜 𝚘𝚗 𝚍𝚎𝚌𝚎𝚗𝚝𝚛𝚊𝚕𝚒𝚣𝚎𝚍, 𝚐𝚞𝚊𝚛𝚍𝚛𝚊𝚒𝚕-𝚏𝚛𝚎𝚎 𝚠𝚑𝚎𝚗 𝚒𝚝 𝚌𝚘𝚖𝚎𝚜 𝚍𝚘𝚠𝚗 𝚝𝚘 𝙽𝚂𝙵𝚆 𝚊𝚗𝚍 𝚛𝚘𝚖𝚊𝚗𝚌𝚎, 𝚗𝚊𝚝𝚒𝚟𝚎 𝚟𝚘𝚒𝚌𝚎 𝚙𝚛𝚘𝚌𝚎𝚜𝚜𝚒𝚗𝚐, 𝚠𝚑𝚒𝚕𝚎 𝚂𝚎𝚜𝚊𝚖𝚎 𝚎𝚖𝚙𝚑𝚊𝚜𝚒𝚣𝚎𝚜 𝚌𝚘𝚛𝚙𝚘𝚛𝚊𝚝𝚎 𝚌𝚘𝚖𝚙𝚕𝚒𝚊𝚗𝚌𝚎 𝚊𝚗𝚍 𝚌𝚘𝚗𝚝𝚛𝚘𝚕𝚕𝚎𝚍 𝚛𝚎𝚜𝚙𝚘𝚗𝚜𝚎 𝚙𝚒𝚙𝚎𝚕𝚒𝚗𝚎𝚜．
You will need a good PC and some technical skill to get this running locally. Hopefully Dorsey also has plans for an app that allows at least the romantic roleplay. In any case, it looks like Maya now has a younger, more attractive, and less frigid love rival to contend with.
