Thinking Machines Labs is working on an interaction model
After months of secrecy, the start-up Thinking Machines Labs is revealing its plans. The company is working on an interaction model.
(Image: photoschmidt / Shutterstock.com)
At the end of 2024, Mira Murati left OpenAI. The departure of the then CTO caused quite a stir in the AI industry. The same applies to the announcement that Murati intends to launch her own start-up called Thinking Machines Labs. The announcement brought the young company more than 12 billion US dollars from investors in a very short time – and that without having a product to show.
Interaction Model from Thinking Machines Labs
This is now changing with a press release from Thinking Machines Labs. The start-up is working on a so-called interaction model to deliver a “scalable approach to collaboration between AI and humans.” In essence, this is an AI that can react to speech in real-time. With response times of around 0.4 seconds, it is intended to enable real dialogues.
These short response times result in several features that other AI models cannot offer. Normally, an AI would let you finish speaking before it answers. Thinking Machines' model is even supposed to be able to interrupt, even if users haven't finished speaking yet. This is useful, for example, if someone has made a mistake and the AI can point it out early. Or it interrupts because it already has an answer ready for you and doesn't need further instructions.
Videos by heise
Another area of application is live translation. The model can listen to the speaker and translate the spoken words simultaneously while they are still speaking. It is also possible to continue talking to the AI while it is working on a previous prompt in the background – for example, visualizing data. According to Thinking Machines Labs, the results of the earlier prompt are then naturally woven into the conversation by the AI.
Thinking Machines Labs has now demonstrated all of its capabilities in a series of short clips. Nevertheless, the model is apparently not yet ready for the public. In the coming months, the start-up plans to release a research preview version to give a limited number of testers access to the AI. Based on the feedback collected, the model is to be further improved.
(mki)
