Skip to main content

Empowering professionals to work smarter


Faster at training models powering an AI assistant


Less compute costs to train models

Otherside AI
SOLUTION: Generative AI
PLATFORM USE CASE: Mosaic AI Model Training

“We wanted to work with a company that was willing to get in the trenches with us, and it was clear that your team was the right fit.”

— Matt Shumer, Co-founder and CEO of OthersideAI

Beginning as an AI lab, OthersideAI builds AI models that enable people to get more out of their daily workflows. Their original product, HyperWrite, allowed people to write better and faster, assisting with everything from outlining to editing and summarizing. OthersideAI’s mission has since expanded to support the vision of creating an executive-level AI personal assistant that would be available to any working professional at a fraction of the normal cost of a human assistant. To take their AI models to the next iteration and build an assistant that could be personalized to each user, OthersideAI needed a machine learning (ML) vendor that could help them optimize every layer of the model development process. For OthersideAI, Mosaic was that vendor, helping their team train AI models 3x faster at 3x less of the cost.

Growing the company’s offerings beyond a flagship product

OthersideAI, a generative AI platform, has always aimed to revolutionize the way AI models are utilized in everyday work. From their inception, the software company focused on creating applications that enhance productivity and efficiency in the workplace. Their flagship product, HyperWrite, exemplified this by offering enhanced writing capabilities across various platforms, whether it be a document or an email. The broader ambition of OthersideAI was to develop an AI personal assistant accessible to all and capable of performing tasks like operating a web browser, ordering a pizza or scheduling a meeting. However, this ambitious project required advanced AI models, and OthersideAI faced a significant hurdle in balancing the advancement of their AI technology with the constraints of limited resources.

The forward-thinking software company recognized the need to refine their base model to ensure its reliability and success in performing new tasks while continuing to learn over time. In fact, OthersideAI also wanted to make personalization core to their AI assistant product, which would require the AI to learn and adapt to individual user behaviors and preferences. This level of customization and learning capability required a more sophisticated approach to model training, where the AI could not only perform tasks but also understand and adapt to user-specific nuances. As a small team, OthersideAI faced challenges in the complexity of developing an advanced AI personal assistant and ensuring its scalability and accessibility to a diverse user base.

OthersideAI had to find a way to efficiently train their AI models without compromising on quality or overextending the team’s resources. This dilemma was exacerbated by the rapidly evolving nature of AI technology, where staying ahead of the curve is crucial for maintaining a competitive edge. OthersideAI found themselves at a crossroads, needing to innovate in AI model training and management while grappling with the practical limitations of a growing but still resource-constrained company. Matt Shumer, Co-founder and CEO of OthersideAI, explained “Although we’ve trained models from scratch, our objective was to end up with the best models at the lowest cost. If we worked with a vendor, we could afford to experiment as we improved our techniques, fine-tuned our datasets and improved upon our models.”

Accelerating innovation with a collaborative approach

OthersideAI’s respect for a hands-on, responsive approach drove their choice to collaborate with Mosaic, which was acquired by Databricks. Unlike the prolonged, sales-driven interactions common with other companies, Mosaic distinguished themselves by their readiness to engage practically and swiftly. The team’s proactive attitude — evidenced by their willingness to initiate a training run within a week — signaled a partnership style that was action oriented and results focused. This moment was pivotal for OthersideAI’s decision on their ML vendor, as it was clear to OthersideAI that Mosaic's strategy was focused on fostering a working relationship with their customers that would help them quickly develop solutions and overcome technical obstacles.

The collaboration with Mosaic allowed OthersideAI to experiment with training models from scratch while fine-tuning base models. Better yet, the training experience on Mosaic's platform was easy to approach. For teams with limited large-scale model training experience, the platform's user-friendly design and ability to strip away the complexities of model training were key highlights. This ease of use enabled OthersideAI to focus on crucial aspects of model training, like data system construction and architectural decision-making, rather than the intricacies of coding.

The proactive support from Mosaic's engineering team, especially in overcoming technical obstacles, further heightened the value of the relationship for OthersideAI. Shumer reiterated, “Mosaic’s engineering team was willing to jump on calls and get us up and running when there were blockers, which was helpful. That's how we ended up achieving milestones so quickly — they could support new models that came out within hours of them being released.” By leveraging Mosaic's advanced tools and expertise, OthersideAI was able to significantly enhance the sophistication and capabilities of their AI models. This was particularly evident in the more complex models, where OthersideAI continues to integrate new features and functionalities. With Databricks and Mosaic, they can now easily fine-tune models for a specific use case, and continue to train it to be smarter and more accurate.

Building out advanced AI models with low overhead

By using Mosaic’s AI training platform, OthersideAI observed a notable decrease in both the time and financial investment required for model development. This economic efficiency, coupled with the improved model capabilities, provided OthersideAI with a competitive edge. Furthermore, the optimized performance, demonstrated by training speeds up to 3x faster than other environments, was a significant advantage. The ability to train models ranging from 1B to nearly 100B parameters with such speed and adeptness saved time and substantially reduced training costs. Shumer elaborated, “Using Mosaic, we can train three models in the time it would normally take us to train one. Because it trains three times faster, we're able to try more things for the same amount of money, which is amazing.”

The ability to rapidly develop and deploy advanced models without incurring prohibitive costs enabled OthersideAI to stay ahead in the fast-paced AI industry, where technological advancements and market demands evolve continuously. The partnership with Mosaic not only streamlined OthersideAI's developmental processes but also empowered the company to embark on more ambitious AI projects, as they continue to build out the AI assistant’s personalization core and further invest in research and development.