MosaicML announces right advance at right time—high-performance, affordable, secure, explainable AI

By Matt Ocko 05.03.23

We have been enthusiastic investors in MosaicML from the start: beginning in 2020, we co-led the company’s first investments on the strength of our belief in CEO Dr. Naveen Rao (above right) and his fellow cofounders like Dr. Hanlin Tang (above left) and key advisors like Dr. Michael Carbin of MIT, who brought profound technical and strategic acumen to a clear goal: “to reduce the time, cost, energy, and carbon impact of AI/ML training so that better AI models and applications can be developed.” Naveen, whom DCVC backed as the Series A lead and largest investor in his ground-breaking AI chip company, Nervana (acquired by Intel), has almost two decades of unique insight into all aspects of the AI space.

Since 2020, two things have happened: ChatGPT and other big-tech large language models have taken the world by storm, inspiring equal parts wonder and worry, and Mosaic has become the leading provider of generative AI infrastructure. Today, building on that strength, the company announced two new sets of products—MosaicML Inference, and a foundation series of models—that will allow developers to deploy generative AI models quickly, easily, and at one-fifteenth the price of other comparable services. Critically, these end-to-end products will allow organizations to train and control large language models in their own secure environments, at multiples of better cost and speed, and to be more likely to achieve alignment and explainability.

This matters tremendously, and right now. One of the biggest threats posed by AI’s explosive current growth is the very real possibility of a dark, self-reinforcing cycle whereby the expense of training AI models leads to the concentration of power in the hands of a few giant tech corporations, maybe a few favored start-ups in their orbit, and the ultra-wealthy. Such a concentration could lead to AI progressively becoming less explainable, understandable, and answerable to key stakeholders, from commercial users, to governments, to individual citizens. The near-term threat of AI isn’t AGI or “paperclips,” but a host of unexplainable, unanswerable AI “local deities” run on expensive, fragile, over-concentrated models whose mistakes are opaque and that disempower both commercial customers and ordinary citizens.

Now is exactly the right time for transparent, customer-controlled, secure AI to become affordable and high-performance. With AI training made dramatically faster and less expensive, it becomes practical for customers to deploy advanced models while understanding AI behavior and aligning it with stakeholder goals. Combined with MosaicML’s ability for people and organizations to have secure and transparent local control over models, Mosaic ensures that AI works for all of us, and safely. Given not only AI’s risks, but also its staggering potential for good—it can be the difference in finding the cure for a disease, in fighting climate change, and in making scientific breakthroughs we can’t yet imagine—its wise and fair deployment is an imperative of historic proportions.

We couldn’t be prouder of the Mosaic team—for being ahead of its time, just in time.

Matt Ocko is Co-Founder and Co-Managing Partner of DCVC.