Creating structure with generative AI

🌈 Abstract

The article discusses the potential of using generative AI to create structure out of messy data, particularly in the context of journalism. It provides several examples of how journalists have used generative AI to summarize complex information, extract key data, and make information more accessible to the public.

🙋 Q&A

[01] Creating Structure with Generative AI

1. What are some examples of how journalists have used generative AI to create structure out of messy data?

• The Marshall Project's banned-books database that Andrew Rodriguez Calderón will talk about in more detail
• A custom GPT by Jaemark Tordecilla that parses complex government audit reports to help expose corruption in the Philippines
• A product recommendation site generated by intelligently mining links in the archives of the newsletter "Why is this interesting?"

2. What is the common element in these generative AI projects? The common element is that they are not really generating something new, but rather creating summaries, extracting information, and structuring data in a more usable form.

3. How did Vikram Oberoi use generative AI to make New York City Council meetings more accessible? Oberoi wrote detailed prompts for GPT-4 to break up the transcript of city council meetings into timestamped question-and-answer pairs, give each chapter a helpful title, and summarize the salient points of the discussion. This created a more navigable and accessible version of the meetings.

4. How did a colleague of the author at The Times use generative AI to extract statistics from New York City Hall press conferences? The colleague gave the transcripts of the press conferences to a large language model (LLM) and asked it to output all the statistics mentioned. This provided a "Harper's Index" of each press conference, which the journalist could then fact-check and use.

5. Why does the author argue that creating structure is the best use case for generative AI? The author argues that generative AI's most powerful use is not in creating entirely new text or images, but in creating structure out of messy data that already exists. This is because AI "hews toward structure" and is less likely to invent information when extracting and organizing data compared to generating text from scratch.

[02] Structuring Receipts

1. How does the author suggest using generative AI to help with the data entry process when submitting expense reports? The author suggests that when traveling for work and accumulating receipts, users can try feeding their receipts to a multimodal large language model and asking for the data they need, such as the amount, date, and vendor. This can help automate the data entry process when submitting expense reports.