Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
Much of the race to own the latest generation of AI is about size—for tech giants rolling out large language models (LLMs), bigger is often better.
The first class of Mozilla’s accelerator program doesn’t share this preoccupation. Instead, the nonprofit org behind the Firefox browser explicitly called for smaller AI models that are able to run locally without the need for cloud servers.
The program recently announced the 14 projects selected, which will each receive up to $100,000 in funding and mentorship from Mozilla. Their use cases range from drug discovery in the Global South and a Swahili LLM to a coding agent assistant and a data tool for artists.
Unlike other Silicon Valley startup programs, the organizers weren’t necessarily looking for projects that could scale into big businesses with VC-sized returns, according to Mozilla Ecosystem Development Lead Liv Erickson.
“We’re really looking to spark innovation in the broader open-source AI ecosystem,” Erickson told Tech Brew. “Our cohort is a mix of for-profit ventures that utilize and contribute back to open-source to community-driven open-source projects and nonprofits that are all working on technology within the open-source and local AI space for the evaluation process.”
Among the picks are French startup Pleias’s ScholasticAI, which uses open-source AI to process scientific documents across different languages; a Koel Labs project that helps language learners improve their pronunciation; and a toolset from a team called Transformer Lab that aims to simplify working with LLMs.
Erickson said the goal of the accelerator is to contribute to a healthy and robust ecosystem around open-source AI. In accordance with the theme of local AI, the models should be able to run on consumer-grade devices like phones or laptops, edge devices, or even private clouds, rather than relying solely on cloud-based APIs, Erickson said.
“We have kind of a broad purview on it, but are really looking at technology stacks that give individual users the ability to understand what they’re running, how it’s being built, how it’s being run, and can do so on increasingly more commercially available hardware,” Erickson said.
Small, local language models have gained some popularity as businesses realize they don’t always need a massive system to address every small use case. They also have the advantage of being more secure.
Erickson said Mozilla is working on coaching each team on their specific goals, which can range from finding the right product market fit to growing an audience or raising funding. Mozilla has also already begun to consider themes for next year’s class of AI builders.
“[It] will likely continue around the theme of democratizing access to these tools and these capabilities and making it easier for a wider audience to use them in a way that preserves their privacy and safety,” Erickson said.