AI Hot 100 Spotlight: - Arcee AI - The Small Language Model Coup D’etat - Overthrowing LLMs
Introducing Arcee AI - one of the Hot 100 startups selected by The AI Furnace, with Brian Benedict (Co-founder) speaking at The AI Hot 100 Conference.
Funding: Arcee AI raised $29.5M.
Founders: Mark McQuade, CEO (ex-Business Development Lead at Hugging Face, ex-Manager at Onica), Jacob Solawetz, CTO (Founding Engineer at Roboflow, ex-research Engineer at Clinic,inc and Founding Research Partner at Qu Capital) and Brian Benedict, CRO (ex-Head of Global Sales at Hugging Face and Tecton)
What does Arcee do? Arcee AI enables enterprises to create their own domain-specific Small Language Models (SLMs), using their own data – empowering them to go from dataset to deployment in one easy end-to-end platform. The startup has pioneered two of the most cutting-edge model training techniques (Model Merging and Spectrum), enabling their platform to create SLMs that outperform traditional closed source Large Language Models (LLMs) on efficiency, accuracy, and all key benchmarks.
Why? SLMs can be highly accurate for specific domains and extremely compute-efficient. Arcee AI mentioned that their models outperform LLMs such as Anthropic and OpenAI when put in production for enterprise clients. Arcee AI mentioned that their customers get excited by the high quality of the SLMs, by the dramatic savings on compute and deployment costs, and by the ease-of-use of the end-to-end platform for training and deploying
Mission: To make custom Small Language Models available to all. Arcee AI envisions a future in which even small companies all have their own SLM, and in which enterprises have many task-specific and topic specific SLMs working at various levels of their organization.
Want to see Arcee AI in action? See Brian Benedict (Co-Founder) speak at The AI Hot 100 Conference.
Q&A with Brian Benedict, Co-Founder
Did you always know you wanted to be an entrepreneur? Were there any other businesses you worked on in the past?
No, in fact I said many times I never wanted to be a founder. But then after doing so many start-ups and being very early on it became clear that the natural progression was for me to go all in on an idea that I was passionate about and Arcee was clearly the right approach in the world of language models for helping businesses achieve their goals.
Can you tell us more about Arcee AI and your perspective on small language models (SLMs) as if you were speaking to the uninitiated? What was the inspiration behind it?
Mark and I worked at HuggingFace where we watched so many companies struggle to get off the ground with LLMs and the pain associated with them when in production. We wanted to build a better system that would allow them to be able to scale LLMs in production - our research pointed us to building Smaller models that were domain adaptive with the companies data. This allowed companies to use their data and for their data to be the primary data source for fueling their language model providing more accurate results. We have never seen a model perform better on tax research when loaded with data on rap music and french poetry.
How did you meet your co-founder? Do you have any advice for entrepreneurs looking for a co-founder?
Mark and I were easy as we worked together and had a shared vision. Jacob and Mark worked at Roboflow together - so the power of working together and getting to know people overtime and having a relationship was very evident and easy for us to transition into being founders together.
What are some brutal truths about entrepreneurship that you wish you knew before starting?
Everything is on you. At the end of the day you want to build a team around you to provide you leverage and there is only so much time in the day but you also have to own way more than you will have ever in any other role you will have in an organization. Hiring across functions becomes even more of a necessity and I always am thinking of how much time I am spending outside my core revenue responsibility to the business and how do I build better company (not just my team’s) health.
There's been a huge interest in agents and agentic workflows. How is Arcee AI thinking about the importance of SLMs in that world?
Agents will continue to be adopted as they drive outcomes on top of the models that are being built. We see agents continuing the growth- but just like the model is only as good as its data. The agent is only as good as its model and we believe building the best models will lead to better agents.
Who’s going to find the most value in what you’re building with Arcee AI? Can you give us examples of some of the use cases that have made the most sense?
We are seeing a myriad of use cases across the board from education, tech, financial services, customer success and others. One of my favorites is match making - meaning finding a person or skill within an organization easier using an SLM that has all of a large organization's hires and when you need an “expert” who are the top recommendations of people to call on for an issue. It has worked now in multiple companies and I feel like it should be more broadly adopted.
What technological advancements have enabled Arcee AI to achieve such high accuracy and compute-efficiency in SLMs?
It all starts with model merging - our ability to merge models and this is our core unique training technique that has enabled our success and how we think about cost optimization coupled with higher performance gains.
How do you see the role of SLMs evolving in enterprise over the next five years?
I think that we will see more adoption of domain specific SLMs from many enterprises who want to move away from hallucinations and drive a more cost optimized solution. The other area that SLMs address with Arcee is ownership - you can build your models and own them which for enterprises gives them even more control.
What is the most surprising thing you found out about the niche you’ve gone into?
There is no loyalty to gen AI vendors - everyone is in the same boat trying to figure out which vendor or method works - this is a good thing by the way.
The AI industry is moving super fast. As a founder, how do you keep up with this for your startup?
Constant reading and conversations with other founders, our friends at HuggingFace are always a great source of information and the constant barrage of Linkedin always helps.
How can The AI Furnace community support you?
We think community is critical and having the ability to engage both in-person and networking with like minded people interested in the space is critical. We love the work The AI Furnace is doing to support start-ups and practitioners looking to do more in this space.
