The four problems to solve

The race to build SportsbizGPT is happening downstream of the bigger fight being waged by the foundational models.

There are four problems to be solved, and the company which solves each of them will be ‘generational’. That's how Jonathon Ross, CEO of Groq sees it.

The four problems represent the evolutionary stages that AI companies must solve to reach full potential:

1. The Hallucination Problem: This is the first and most critical hurdle to overcome. It involves ensuring that models do not generate false or fabricated information, which is a foundational requirement before more complex tasks can be reliably automated.
2. Breaking Down Sub-goals for Agentic AI: Once hallucinations are addressed, the next challenge is enabling AI to effectively deconstruct large objectives into smaller, actionable sub-goals. Solving the hallucination problem first is vital here because long chains of agentic tasks are currently prone to introducing errors and "hallucinations" that can derail the entire process.
3. The "Invent" Stage: Currently, Large Language Models (LLMs) function by making the most probable prediction, which results in predictable and often "terrible" creative output in art and writing. The challenge for the "invent" stage is to move beyond probability to generate insights or content that are non-obvious yet clearly correct once they are seen.
4. The Proxy Stage: The final stage is reached when a model can effectively proxy decisions for the user. This involves the model acting with the level of trust and authority granted to a Chief of Staff or Executive Assistant, making autonomous choices regarding logistics, scheduling, and priorities—such as deciding which interviews to take and which flights to book.

This progression is similar to climbing a ladder where each rung must be secure before reaching for the next; you cannot trust an AI "Proxy" to handle your entire schedule (the top rung) if it still "hallucinates" your destination or fails to understand the "sub-goals" of travel logistics (the bottom rungs).

Jonathan Ross’s third bullet point is the most interesting and relevant to the SportsbizGPT question. 

The Invent Stage is about the machine providing non-obvious answers.

From here you get to the ‘proxy stage’, when you can trust an LLM to make decisions on your behalf. Because making decisions is a creative act. It requires selecting one option and forgoing others. 

Each of the four problems will be solved by ‘an industry defining tech company’. That’s the big race. And it will set off a series of other races. 

Who will build the machine that makes real sense of sports media, betting, ticketing and fan engagement datapoints? 

Whoever it is, they’ll be rich.