AI startup Underdog has released a game demo built with its new text-to-speech model, Woof, which is 31.7 MB. The demo works in a browser, and the company says the model can run on most devices without an internet connection.
The game, shared by the company on X, presents a retro tech-themed room where users can type on a typewriter and hear the text spoken back, or rotate a tuner to generate music. Underdog says the model has 31 million parameters and is open for anyone to download.
The demo comes weeks after the company launched Bark, a 167 MB speech-to-text model. Underdog has also released a 27-billion-parameter model that can run on laptops, and has a waitlisted app it positions as a local alternative to products such as Muse and Instinct.
Sigil Wen, Underdog's founder, said he wanted to build a system with AI models that run on users' devices and let them own their data. "I think we will be able to run most capable models of today on the devices you already own within months. We just want to train the best open models that are American-made, that are open, that are capable and allow people to have private AI," he told TechCrunch, referring to his manifesto.
The game's intro warns players about "intrusive" cloud providers, ostensibly a jab at cloud-based models.
Analysis
Why This Matters
- Demonstrates that capable AI models can run locally, reducing reliance on cloud servers and improving privacy.
- Opens the door for voice assistants, personalised podcasts, and other applications on consumer devices without internet.
- Suggests a growing trend of 'edge AI' competing with massive cloud models.
Background
The race in AI has focused on ever-larger models requiring huge compute and cloud infrastructure. A counter-movement of startups and researchers is creating smaller models that can run on edge devices like phones and laptops, aiming to offer privacy and offline capability.
What to Watch
- Whether Underdog's model achieves real-world adoption for local assistant experiences.
- How larger AI companies respond with their own smaller models.
- The success of Underdog's waitlisted app as a local alternative to cloud-based tools.
Checked
12 claims in this story checked against 1 source: 12 supported.
Show each claim
- SupportedUnderdog released a 32MB voice model that runs in a browser.TechCrunch: “One such startup, Underdog, has released a game built with its new voice model called Woof that’s only 32MB large.”
- SupportedThe game demo showcases text-to-speech and music generation using a small open model.TechCrunch: “You can type anything on the typewriter using your keyboard, and the game will speak it back. You can also rotate the tuner to generate different kinds of music. ... The model has 31 million parameters and is open for everyone to download.”
- SupportedThe game uses Underdog’s new text-to-speech model Woof, which is 31.7 MB.TechCrunch: “We made a game for our new frontier text-to-speech model, Woof. 31.7 MB, so small that it runs in your browser”
- SupportedThe demo works in a browser, and Underdog says the model can run on most devices without an internet connection.TechCrunch: “All of this works within the browser through Underdog’s speech generation model, which, the company says, can run on most devices without any internet connection.”
- SupportedThe company shared the game on X; users can type text to hear it spoken or rotate a tuner to generate music.TechCrunch: “Play the game on X… all voices are Woof. Defeat Darkcloud for an invite”
- SupportedWoof has 31 million parameters and is open for anyone to download.TechCrunch: “The model has 31 million parameters and is open for everyone to download.”
- SupportedThe demo came weeks after Underdog launched Bark, a 167 MB speech-to-text model.TechCrunch: “This demo comes weeks after Underdog launched Bark , a 167 MB speech-to-text model”
- SupportedUnderdog has released a 27-billion-parameter model that can run on laptops.TechCrunch: “Underdog has also released a 27-billion-parameter model that can run on laptops”
- SupportedUnderdog has a waitlisted app positioned as a local alternative to Muse and Instinct.TechCrunch: “along with a waitlisted app that the company is positioning as a locally-hosted alternative to Muse and Instinct.”
- SupportedSigil Wen, Underdog’s founder, said he wants to build a system with on-device AI models that let users own their data.TechCrunch: “Sigil Wen, the founder of Underdog , told TechCrunch that he started building the app because, instead of using notetakers like Granola, he wanted to build his own system with AI models that can run on users’ devices and let them own their data.”
- SupportedWen told TechCrunch: “I think we will be able to run most capable models of today on the devices you already own within months. We just want to train the best open models that are American-made, that are open, that are capable and allow people to have private AI,” referring to his manifesto.TechCrunch: ““I think we will be able to run most capable models of today on the devices you already own within months. We just want to train the best open models that are American-made, that are open, that are capable and allow people to have private AI,” he said, referring to his manifesto .”
- SupportedThe game’s intro warns players about “intrusive” cloud providers, ostensibly taking a jab at cloud-based models.TechCrunch: “The intro warns users about “intrusive” cloud providers in the town — ostensibly a jab at cloud-based models.”
