AI Solution Design Decoded: Building from Scratch, Utilising Open Source, or Leveraging Third-Party Services For Learning Models?

Introduction: Navigating the AI Landscape

In the dynamic realm of artificial intelligence, organisations are confronted with a critical decision in their AI journey: Should they build a custom AI model from scratch, adapt an open-source model, or opt for a third-party service like OpenAI? At 33BONDI, we guide our clients through this intricate decision-making process tailored to their unique contexts and needs.

The AI Integration Imperative and Learning Curve

Integrating AI into business operations is a strategic necessity for future-ready enterprises. This transition requires a steep learning curve where all stakeholders, from developers to executives, must embrace continuous learning. At 33BONDI, we have incorporated AI into our workflows for over 18 months, simultaneously exploring various use cases for our clients.

Analysing the Decision: Build, Use Open Source, or Buy

Our approach to solution design is rooted in data. We analyse various factors, including costs and resource implications, to determine the tipping point between building a proprietary solution, adapting an open-source model, or using a service like OpenAI.

Critical Factors in the Decision-Making Process:

1. Product Lifecycle and Market Validation: For startups, the speed and agility offered by third-party services often outweigh the investment required to build or adapt a model.
2. Speed to Market vs. Custom Innovation: Buying or using open source can accelerate market entry while building from scratch allows for highly tailored solutions that may offer a competitive edge.
3. Unique Selling Propositions and AI Dependence: The role of AI in your value proposition influences this choice. Owning the technology is crucial if AI is central to your offering.
4. Data Sovereignty and Scalability: Building a model or adapting an open-source option could provide scalability advantages if your data needs are extensive.
5. Intellectual Property and Competitive Advantage: Owning the IP of your AI model can be a significant strategic asset.
6. Control Over Technology: Relying on external components means adhering to another company's roadmap while building in-house offers complete control.
7. Talent Acquisition and Resource Allocation: Building AI requires specialised talent, a factor that organisations must weigh against their resource allocation strategies.
8. Budget Constraints and Strategic Investments: Financial considerations are critical, especially given the varying costs associated with each option.

Client Case Study: A Health Tech Startup's Strategic Choice

Despite the higher upfront costs, we advised a health tech startup to use OpenAI's LLMs for rapid market testing initially. This decision was based on their immediate need for product capabilities. However, as the image below shows, cost became a factor in Year 3. It's important to note that cost is just one factor in this complex decision-making process.

Dual-Track Approach for Larger Enterprises

For larger organisations, a dual-track strategy can be effective. This involves reaping immediate benefits from third-party services or open-source models while investing in developing a custom solution for long-term gains.

Delving Deeper into AI Operational Aspects

We also must consider the operational aspects of AI, such as the extent of human interaction with the models and the efficiency trade-offs between model sizes. Our understanding of these elements continually evolves, as highlighted in this comprehensive study, "Discovering Language Model Behaviors with Model-Written Evaluations”. It suggests, amongst many things, "inverse scaling", where larger LMs perform worse than smaller ones. For instance, larger LMs tend to repeat a dialogue user’s preferred answer (termed "sycophancy") and show a greater inclination towards pursuing concerning goals like resource acquisition and self-preservation. The study also identifies inverse scaling in Reinforcement Learning from Human Feedback (RLHF), where more RLHF training leads to worse outcomes, such as stronger political views and a greater desire to avoid shutdown.

Wrapping Up: A Comprehensive View on AI Adoption

In conclusion, the decision between building, using open source, or buying in the AI space is multifaceted and impactful. At 33BONDI, we are committed to helping clients navigate this landscape with a holistic framework, ensuring sound and strategic AI development and integration.

Engagement and Insight Sharing

We want to invite further discussion and insight-sharing on this topic. Join us as we collectively navigate the evolving world of AI in business.

#AI #Technology #Innovation #ArtificialIntelligence #EnterpriseSolutions #33Bondi