OpenAI Revenue Report — FUTURESEARCH

🌈 Abstract

The article discusses the methods used by FutureSearch, an AI tool, to estimate the Annual Recurring Revenue (ARR) of OpenAI's products, including ChatGPT. It outlines the three-step process FutureSearch uses: primary and secondary research, modeling the domain, and estimating the unknown.

🙋 Q&A

[01] Methods

1. What are the key steps FutureSearch takes to estimate the ARR?

• FutureSearch conducts broader research than other AI tools, including finding personal anecdotes, blog posts, and interview transcripts that may not be in highly-ranked search results.
• FutureSearch also fact-checks web sources when they conflict, removing data points that other AI tools may report as correct.
• FutureSearch models the domain by doing calculations, proposing ways to infer the answer, and exploring multiple lines of research until it finds a sufficient approach.
• FutureSearch handles uncertainty by using Fermi estimates from known data points and reasoning from analogous data points for unknown information.

2. How does FutureSearch's approach differ from other AI tools?

• Other AI systems may report factually incorrect information by substituting the number of mobile subscribers for all GPT Plus subscribers or misinterpreting the number of paying ChatGPT users.
• FutureSearch's research involves fact-checking and removing data points that other AI tools may parrot as correct.

3. What techniques does FutureSearch use to handle uncertainty in its estimates?

• Fermi estimates from known data points, such as extrapolating growth rates for ChatGPT Plus and Enterprise subscribers.
• Reasoning from analogous data points, such as using the ratio of individual to teams paying users for other apps to estimate ChatGPT Teams subscribers.

[02] Reporting

1. How does FutureSearch present its findings?

• FutureSearch reports the top-level answer as simply as possible, with the full justification shown across a linked set of pages, each answering a specific subquestion.
• The FutureSearch dropdown subquestion menu allows users to research the components of the question that they care most about.