‍Gregor: You are building an interesting early stage startup using AI and human reinforcement. In addition, you are the chairman of Yseop, which has been building a NLG platform for over 10 years. Can you talk about how to improve generative AI systems so users can achieve their goals in a seamless way? How important is the UX of the customer journey in developing these future product roadmap?

‍

Furthermore, it is fundamental to use industry specific ontologies

‍

Craig: The use of ChatGPT/Bing and GPT4 seems pretty simple for technologists, but to someone with less tech confidence, I can imagine that the process and capabilities haven't been adequately explained. As we all know, thoughtful and efficient UX makes everything more useful, so the importance of user experience design cannot be understated. But these tools still have areas where they must progress. It will be key to reach 100% accuracy with the help of real-time user feedback loops and RLHF (which AI Redefined and OpenAi provide, more details in my LinkedIn post). Furthermore, it is fundamental to use industry specific ontologies, like we use at Yseop, to generate language that meets the quality standards of a specific industry. Most large corporations have disallowed the use of ChatGPT because of data leakage as these solutions don't silo users data; a sophisticated user can extrapolate sensitive data based on user queries. A few major issues need to be fixed before this is a real solution for most enterprises.

‍

Gregor: We know that AI learning models are based on datasets annotated or created by humans, and can further be improved based on insights from humans to improve the accuracy of a generative AI system over time. Do you have any tips on how to build a business model that includes the cost of R&D and the iterative improvements of models when you architect an AI product experience?

‍

Craig: I think 'usage-based pricing' (similar to cloud or CPU pricing) is necessary for these systems. We developed an open source framework called Cogment that is designed to enable Humans and AIs to operate in shared environments. It is designed to allow the training of complex agent architectures on sequential decision-making tasks in complex environments and supports humans-in-the-loop. Building on frameworks like these can enable innovative product solutions in many new applications.

‍

‍

Gregor: AI technology is destined to be a $200B industry in 2030, 7 short years away. Do you think that the current goldrush is overstated, or can generative AI systems fundamentally redefine the product roadmap in many industries? What are the roadblocks to make the technology leap feasible across many organizations?

‍

Craig: Generative AI will cause a fundamental change to business because authenticity, accuracy, data privacy, IP ownership, specific ontologies will all impede progress. It will rapidly become our personal assistant and search engine, but also can change many industries if these roadblocks are solved, which might not happen linearly or incrementally. We are currently building AI systems that help Humans in high stress situations by increasing the cognitive load on their critical tasks and I believe that these industry-specific applications can augment humans daily workload exponentially. I encourage everyone to listen to the Sam Altman/Lex Fridman podcast which covers a wide range of topics around UX, AGI and the impact on the future of work.

‍

Gregor: Thanks Craig for those amazing insights!