AI voice chats often feel uncomfortable because the assistants struggle to determine when to speak.

      Thinking Machines Lab has announced that it is developing full duplex AI, which allows an AI system to listen to what a person is saying while simultaneously generating a response. In simpler terms, this means it functions more like a phone call rather than a walkie-talkie.

      Founded last year by former OpenAI CTO Mira Murati, the startup introduced interaction models, beginning with TML-Interaction-Small. They claim that the system can reply in 0.40 seconds, which is comparable to typical human conversation speeds.

      However, there is a limitation for those wanting to try it out immediately. This is currently a research preview, with limited access planned for the next few months and a wider release anticipated later this year.

      A quicker type of AI exchange

      The underlying concept is straightforward and significant. Rather than waiting for someone to complete their statement before formulating a response, the model processes incoming speech while simultaneously crafting its reply.

      This reduction in delay is important because pauses can make AI assistants seem unnatural. Thinking Machines Lab presents TML-Interaction-Small's response time of 0.40 seconds as nearly akin to natural conversational pace, which could signify a significant improvement for voice technologies.

      They also assert that this speed surpasses similar offerings from OpenAI and Google. While this benchmark lends credibility to the announcement, external users still need to verify whether the interaction feels as seamless as the figures suggest.

      When speed influences interaction

      An assistant that responds while still gathering information alters user expectations for voice conversations. While dialogue can progress more quickly, the system must also carefully manage timing.

      This balance is critical when someone seeks a quick clarification instead of a lengthy response. Rapid replies are of little value if the assistant interjects prematurely, misinterprets the speaker, or disrupts the flow it’s meant to enhance.

      For now, the architecture is the main highlight. The true test of the product will be whether the interaction model can create a natural timing experience.

      Key points to observe before launch

      The launch timeline is a crucial aspect at this point. Thinking Machines Lab has indicated that a limited research preview will be available in the coming months, followed by wider access later this year.

      Details regarding availability, pricing, supported platforms, and performance outside of controlled environments remain uncertain. These factors are significant because a faster model will only be beneficial if users can integrate it into their everyday voice tools.

      For those who utilize AI voice assistants, the best course of action is to monitor the preview closely. Full duplex AI holds promise, but practical testing will determine whether quicker responses genuinely enhance routine AI conversations.