DeepSummary
Mark Hoang, co-founder of Gradient AI, discusses the origins of the company and its focus on building automation for enterprise workflows using AI. The company initially targeted knowledge workers and analytics teams before pivoting to target enterprises in finance and healthcare verticals with agentic automation solutions. Gradient recently open-sourced a version of the Llama 3 model with extended context length to enable new use cases.
Hoang shares insights on the challenges of positioning and selling an AI product to enterprises, emphasizing the importance of understanding organizational dynamics, stakeholders, and resonating with users' problems. He stresses the need for constant alignment within the startup team and clarity on value proposition amidst market noise.
The conversation explores key model advancements in the past year, including improved instruction following, longer context lengths, and multimodality. Hoang highlights the trade-offs between longer contexts and latency/throughput, and the potential benefits for tasks like observability and understanding across disparate data sources.
Key Episodes Takeaways
- Gradient AI started by targeting knowledge workers and analytics teams before pivoting to focus on enterprise clients in finance and healthcare verticals.
- They open-sourced an extended context version of the Llama 3 model to enable new use cases and engage the community.
- Key challenges included positioning their AI solution amidst market noise and understanding organizational dynamics for enterprise sales.
- Major AI advancements in the past year included improved instruction following, longer context lengths, and multimodality.
- Hoang emphasizes the importance of constant team alignment, understanding the customer's workflow and who is impacted for successful enterprise AI adoption.
- Trade-offs exist between longer context lengths enabling better understanding across disparate data sources and increased latency/throughput.
- Open sourcing allows engaging the community while keeping enterprise-focused offerings proprietary with support.
- Selling to enterprises requires relentless efforts to understand stakeholders, organizational impacts, and resonating the value proposition.
Top Episodes Quotes
- “Within selling into the enterprise. For AI, you got to know who's going to be using your product and who it impacts, because without that, you're sitting in this isolated bubble where you think that as long as this person's on board, I can sell my product and I know it solves his problem. But who does solving his problem? Do you impact the adjacent team in some way and then do you make their life a little bit harder?“ by Mark Hoang
- “For us, when we think about it, it's about what value can we democratize to the community and then what feels like enterprise and what feels like something where people want constant support for.“ by Mark Hoang
- “Part of that is the effects of that, as we kind of talked about already, is like learning how to sell, learning how organizations work. Like, it was apparently very clear to us the only way we can be successful as an enterprise AI company is to try those things.“ by Mark Hoang
Entities
Company
Person
Product
Episode Information
Open Source Startup Podcast
Robby (Cowboy VC) & Tim (Essence VC)
7/1/24
Mark Huang is Co-Founder of Gradient, the platform for enterprise agentic automation. Gradient recently open sourced their 4M context window finetune of Llama-3, which is the longest context window available today.
Gradient has raised $10M from investors including Wing VC, Mango Capital, and Tokyo Black.
In this episode, we dig into enterprise readiness for LLM-backed applications today, Gradient's approach to pushing context lengths for foundation models and the benefits to open sourcing their Llama-3 finetune model, their focus on healthcare and finance verticals, how they're finding their place in the noisy GenAI infra space & more!