Cooperate and Graduate: A Better Way to Work with Generative AIs

My co-founder and I are both graduates of the U.S. Army Ranger School, often considered the hardest course in the U.S. Army. Ranger School was daunting. Our class graduated less than 40% of the candidates who started the course. It was 63 days of sleep and food deprivation and being pushed to and often beyond one’s limits. 

The key to surviving and graduating Ranger School lay in the often repeated mantra “cooperate and graduate.” No Ranger can succeed by themselves. In fact, the very purpose of Ranger School is to teach its graduates how to lead, follow, and work with others under the most difficult of conditions.  

More on this and its relevance to Generative AI in a moment. 

A few weeks ago, we announced we were developing boodleGPT, a middle layer AI to enable better sales, marketing, and fundraising decisions. And we did – we developed the alpha version of an “engine” that would drive boodleGPT. We solved how to bring structured data into an LLM while maintaining privacy and security.  We figured out how to combine a generative AI with our proprietary dataset of 35 billion consumer and philanthropic insights. 

We then put boodleGPT Alpha into the hands of users. Then we asked questions. And listened. 

And while we learned plenty about how boodleGPT Alpha performed (and didn’t perform), we learned even more about areas that went beyond just boodleGPT. Several things stood out: 

1. Users cared less about how powerful an AI Assistant was and more about how personalized it was to their specific wants and needs.

2. As powerful as any single AI Assistant could be, a collaborative team of AI Assistants with different capabilities is better able to do the things that users want out of Generative AI. 

3. Users want to collaborate with other users (and with their teams of AI Assistants). 

We then deep dived into how others were tackling the above issues and found that, by and large, the current focus in generative AI is on making better, more specialized LLMs or building tools to enable others to create better, more specialized LLMs. There just simply isn’t a lot of time and attention being paid to how to enable users to personalize then collaborate with multiple assistants and make that work experience easier, better, and more productive, especially with other users. 

Here lies a challenge and an opportunity:

Create a platform that allows multiple users and multiple generative AI’s to “cooperate and graduate.” 

We believe that how we interact with Generative AI is just as important as what Generative AI can do – in fact those two concepts are inextricably linked. Here’s our list of what we think an ideal user experience with Generative AI should include:  

• Selection. A user should be able to select ANY AI they want for a particular task.  This includes different foundation AI’s and AI’s based on different foundation AI.  If a user wants to use an OpenAI LLM, a Google LLM, and an open source LLM, they should be able to. 
• Specialization/Personalization. A user should be able to specialize/personalize their AI.  This means being able to change the foundation AI, the initialization/personalization, and which first or third party data sets the AI has access to.  
• Security. A user should be able to provide access to proprietary or confidential information to their AI without concerns that it will be shared with any unintended third parties or used for training other LLMs.  
• Switching. A user should be able to easily switch between AI’s in the same platform and even the same chat. 
• Sharing Work Among AI. A user should be able to work with multiple different AI’s at the same time to solve a problem or answer a question.    
• Multi-step Solutions. A user should be able to solve problems that require multiple steps from multiple different AI’s (both through user-specified chaining and through reliable agents). 
• Status Updates. A user should be able to receive updates from their AI Assistants performing tasks, especially those tasks that take a length of time or require multiple steps. 
• Sourcing/Justification. A user should be able to quickly see the sources used by the AI to generate a response and/or a justification for the response. 
• Safety. A user should not be able to generate responses that are unsafe for others. 
• Seeing Answers. A user should be able to see answers in different formats, including visualizations (charts, graphs, video, etc.). 
• Speak and listen. A user should be able to speak a prompt and listen to the response. 
• Saving. A user should be able to save AI generated content. 
• Structure/Organization. A user should be able to structure and organize the AI generated content they save. 
• Sharing and embedding. A user should be able to share and embed their AI generated content and dynamically update it. 
• Socialize. A user should be able to share their AI generated content with a social network and receive feedback. 
• Synchronized Problem Solving/Collaboration. A user should be able to synchronize efforts with other users and their AI teams to collaboratively solve a problem. 
• Searching.  A user should be able to search their own AI generated content as well the AI generated content that others make publicly available. 
• Shopping/Selling. A user should be able to shop for additional AI Assistants and sell AI Assistants to other users. 
• Single Place. A user should be able to do all of the above in a single platform. 
• Simplicity. A user should be able to do all of the above in as simple a fashion as possible. 

Creating a platform that allows multiple users and multiple generative AI’s to “cooperate and graduate”  is just as important – perhaps more important – than developing our own proprietary generative AI Assistant. And we’re building something to do just that.