Dazzle, a new AI startup led by Marissa Mayer, raises $8M – News

Dazzle AI Inc., a startup founded by former Yahoo Inc. Chief Executive Marissa Mayer, today announced that it has raised $8 million in funding.

Prominent venture capitalist Kirsten Green led the investment. She was joined by Mayer, Kleiner Perkins, Greycroft, Offline Ventures, Slow Ventures, Bling Capital, Amino Capital and Acquired Wisdom Fund. The round values Dazzle at $35 million.

Before launching the company, Mayer led Yahoo as CEO for five years. She oversaw the sale of the onetime internet giant’s core assets to Verizon Communications Inc. in 2017. She earlier held senior roles at Google LLC and helped lead the development of the search giant’s advertising services.

According to News, Dazzle is developing an artificial intelligence personal assistant. The company’s website states that its “mission is simple: to bridge the gap between what you want and what they can do with AI.” The website also specifies that the software will be geared toward the consumer market.

Dazzle reportedly emerged from another consumer technology startup, Sunshine, that Mayer founded in 2018. The latter company developed a contact management app with AI features. The app enabled users to deduplicate information in Google Contacts, enrich it with external data and access it through a search bar. Last March, Sunshine added a tool for organizing events.

Mayer shut down the company a few months after the feature update. Sunshine failed to draw a large user base, with the app’s Google Play listing showing that it was only downloaded about 1,000 times. The company’s investors received a 10% stake in Dazzle.

A job posting for an iOS developer on Dazzle’s website indicates that it plans to make its AI assistant available on mobile devices. Dazzle is also seeking an AI researcher who can help its team “architect, develop and evaluate AI models.”

Training large language models can be a highly expensive undertaking. Given Dazzle’s modest initial funding, it’s possible the company will seek to develop a customized version of an existing, open-source LLM instead of training a model from scratch. The former approach is more cost-efficient thanks to model customization methods such as LoRA that reduce infrastructure requirements.

Model training is not the only major expense involved in bringing a consumer AI service to market. Depending on the service’s popularity, processing inference requests can require even more infrastructure. Dazzle might seek to minimize its inference costs by using on-device AI models to power its assistant instead of sending requests to a cloud-hosted model.

The company reportedly plans to share more information about its service early next year.

Image: Unsplash