WPP’s Integration of Anthropic’s Claude 3 Models, Leading Agency AI Interoperability

In the rapidly evolving landscape of generative AI, agencies are increasingly focusing on interoperability between various AI models to develop robust strategies for marketing and other applications. WPP’s recent integration of Anthropic’s Claude 3 model into its AI platform, WPP Open, is a prime example of this trend. This move not only adds Anthropic to the list of rival models from OpenAI, Google, and Stability AI but also highlights the importance of choosing the right AI models for specific use cases.

The Claude 3 Model

Released in March, Claude 3 comes in three versions: Opus, Sonnet, and Haiku. Each model has its own unique capabilities, costs, and performance characteristics.

• Opus, the largest and most expensive, offers the highest level of capability but it’s also the slowest.
• Sonnet, the middle-sized model, strikes a balance between intelligence and speed.
• Haiku, the smallest and cheapest, is the fastest of the three.

WPP’s CTO, Stephan Pretorius, emphasized the benefits of Anthropic’s models in terms of transparency, scalability, guardrails, and flexibility. By using Claude 3 through Amazon’s Bedrock platform, WPP gains access to enterprise-grade AI applications with improved accuracy, security, and scalability.

WPP’s AI Application Strategy

WPP’s approach to AI applications revolves around three key categories: ideation, automation, and optimization. These categories are applied across various domains, including strategy, consulting, brand design, audience personas, content production, CRM applications, and public relations.

Pretorius described WPP’s approach as a “Claude inside strategy,” drawing parallels to Intel’s iconic “Intel inside” tagline and Qualcomm’s recent strategy for its Snapdragon AI chips. He praised Anthropic’s focus on safety, fairness, and the ability to explain and interpret the results generated by their models.

Training AI Models

WPP’s process of training AI models involves incorporating various types of data from across the holding company, including the dozens of advertising, communications, public relations, marketing, technology and eCommerce companies under the brand’s umbrella.

This data is organized into four different “brains” that contain information about audiences, performance, brands, and channels. The data ranges from research and marketing creative to audience data and brand value trackers.

Choosing the Right LLM

Deciding which large language model (LLM) to use requires careful consideration of factors such as authentication, responsibility management, data storage, and knowledge graphs.

While marketing companies initially sought to adopt a single LLM for various tests, they are now increasingly integrating multiple LLMs into their AI platforms based on specific use cases and data storage requirements.

WPP Open aims to bring more interoperability between cloud providers and LLMs through hyperscalers. Pretorius highlighted the significant opportunity to access code via Amazon Bedrock, which aligns with the emerging landscape of AI integration.

Anthropic’s Research on Claude 3

Anthropic recently released new research to help explain the datasets inside the “black box” of Claude 3 and better connect the dots between datasets and AI model outputs.

The research aimed to understand the “mind” of Claude by identifying millions of concepts triggered when it reads texts or sees images.

The explainability of AI has been a long-standing challenge, even for the experts developing the models. This difficulty in understanding how AI models work has hindered efforts to improve them and achieve consistency across various applications.

Startups Tackling AI Explainability

The challenges surrounding AI explainability, accuracy, and copyright concerns have created opportunities for startups to develop new lines of business. Patronus AI, for example, recently announced a $17 million fundraising round to develop ways to audit AI models faster, cheaper, and at scale.

Their tools help check for copyright infringements, hallucinations, and accuracy in real-time based on various prompts and outputs.

Patronus AI CEO and co-founder Anand Kannappan emphasized the importance of these capabilities, stating that they do not exist today due to the expensive and slow nature of developing and evaluating large language models.

WPP’s integration of Anthropic’s Claude 3 model into its AI platform represents a significant step forward in agency AI interoperability. By leveraging the unique capabilities of Claude 3 and incorporating it alongside other rival models, WPP is positioned to develop more robust and effective AI strategies for marketing and other applications.

As agencies continue to navigate the complex landscape of generative AI, the importance of choosing the right AI models and ensuring interoperability between them will only grow. WPP’s approach, combined with the ongoing research and development efforts of companies like Anthropic and startups like Patronus AI, will play a crucial role in shaping the future of AI in marketing and beyond.

---

If you need assistance understanding how to leverage Generative AI in your marketing, advertising, or public relations campaigns, contact us today. In-person and virtual training workshops are available. Or, schedule a session for a comprehensive AI Transformation strategic roadmap to ensure your marketing team utilizes the right GAI tech stack for your needs.