AI on Your Device: Underdog’s New Lightweight Model

While the current trend in artificial intelligence focuses on increasing model size and computational power, the startup Underdog is moving in the opposite direction. The company has introduced a new voice model called Woof, which occupies just 32 megabytes. This allows users to run it on standard devices, such as everyday smartphones or personal computers.

Technology Without the Internet

To demonstrate Woof's capabilities, the company created a special web game. In a retro-style virtual room, users can use a typewriter where typed text is instantly converted into speech. The entire process takes place locally within the browser, without the need for cloud services or an internet connection. This is a significant step for data privacy, as user information never leaves the device.

The model, which features 31 million parameters, is open to the public and available for anyone to download. While it cannot compete with giant models in terms of scale, it is fully sufficient for powering personal assistants or generating podcasts.

Goal: AI for Everyone

Underdog founder Sigil Wen states that the company's primary mission is to give users full control over their own data. Just a few weeks ago, the startup released a 167MB model called Bark, which focuses on speech-to-text conversion. The company's portfolio also includes a more powerful 27-billion parameter model optimized for operation on laptops.

According to Wen, technological development is moving so quickly that in the coming months, users will be able to run the most powerful models on the devices they already own. Underdog's strategy is based on American manufacturing, open source, and the development of personal artificial intelligence.

What This Means for Users

For users who are often dependent on cloud services, this development opens up new perspectives. The advancement of local models means that AI tools will become faster, cheaper, and, most importantly, independent of internet quality or server availability. Although such technologies are still in their early stages, the trend is clear: artificial intelligence is gradually moving from the cloud to our personal devices.