Is Smaller Better?

September 30, 2024

Hi friends,

Lately, I’ve been thinking a lot about Large Language Models. While models like GPT-4 are undeniably impressive, their training and operational costs are incredibly high. In most cases, you don't need access to the entirety of human knowledge to complete a simple task. I believe that for most applications, a highly specialized model designed for your specific problem is far more valuable. These considerations, among others, have led me to place a big bet on Small Language Models.

SLMs offer several key advantages:

  1. Cost-effectiveness: They require less computational power, making them accessible to startups and smaller teams.

  2. Specialization: SLMs can be fine-tuned with precision for specific tasks or industries.

  3. Faster development: Smaller models allow for quicker training and iteration cycles.

  4. Better privacy: Many SLMs can run locally, keeping sensitive data on-premise.

At Flybridge, we’re starting to see a shift towards Agentic AI, where multiple SLMs work together. This approach is more efficient and allows for greater specialization. Soon, we’ll see teams of AI models, each an expert in its domain, collaborating to solve complex problems.

This opens up exciting possibilities:

  • In healthcare, models could specialize in diagnostics, treatment planning, and drug interactions.

  • Fintech companies could leverage models for market analysis, risk assessment, and fraud detection.

  • Educational platforms could adapt to individual learning styles and subject areas.

Our investment in Arcee is a perfect example of this trend. Founded by former Hugging Face execs and led by Mark McQuade, Brian Benedict, and Jacob Solawetz, whom I've known for years, Arcee is building an end-to-end platform for training, deploying, and monitoring Small Specialized Language Models.

What I love most about Arcee is their focus on enterprise AI adoption. They've developed a system that allows companies to host models in their own cloud infrastructure, ensuring data never leaves the organization. This addresses major concerns around security and reliability that we've been hearing from enterprise customers.

I'm eager to meet founders who have unique knowledge or insights in this space. When evaluating startups in this area, I look for deep domain expertise, novel approaches to model design, and a focus on solving real-world problems.

Bigger isn’t always better. In the near future, I believe users will prefer smarter, more efficient, and more specialized systems that work together to solve complex problems.

If you know any founders working on small, specialized AI models, or agentic AI systems, I'd love to chat.

Until next time,

Jesse

🚀 General Collaboration is launching next week and hiring a hiring Reverse Engineer. They’re building Superhuman for comments in apps we use at work (GDocs, Figma, Linear, etc.). Support their launch and apply now.

💼 Arcee is hiring a Platform Engineer. The opportunity is fully remote. If you or someone you know might be a fit, shoot me an email. Apply here.

💰 Next Wave NYC is a pre-seed venture fund run by some of the best and brightest founders and operators in NYC. Contact us.

👋 Hi, I’m Jesse, a seed-stage VC at Flybridge. If you’re building or investing in an AI-driven future, reply to this email with a short message. If you’re local to NYC, I’d love to catch up over coffee.