How The New York Times is Using Generative AI as a Reporting Tool

Overview

• The New York Times (NYT) utilizes generative AI as a tool to aid reporters without replacing them.
• AI assists in handling large volumes of data, allowing reporters to focus on the nuanced aspects of reporting.

Use of AI in Reporting

Transcription

• NYT used AI to transcribe over 400 hours of audio from the Election Integrity Network meetings.
• Automated transcription tools converted audio to text, resulting in nearly five million words.
• AI transcription has improved significantly, with accuracy rates improving from 73% in 2018 to 94% in 2024.
• This process aids reporters in quickly and accurately transcribing audio data at a lower cost.

Analysis

• After transcription, large-language models (LLMs) were used to search for relevant topics, notable guests, and recurring themes in the transcripts.
• LLMs help in summarizing complex documents, but have limitations such as confabulation and lack of deep understanding of context.

Human-AI Collaboration

• Reporters manually reviewed AI-selected passages for accuracy and contextual relevance.
• Human judgment is crucial to ensure that all quotes and video clips are accurate and fairly represent the original context.
• By combining LLM capabilities with human insight, NYT can leverage AI's strengths while mitigating its weaknesses.

Limitations of AI

• Australian government study indicated that AI summaries often lack depth and can be factually inaccurate.
• AI's ability to understand subtle nuances or implicit meanings is limited compared to humans.

Conclusion

• Generative AI serves as an aid rather than a replacement for human reporters.
• AI functions similarly to a "drug-sniffing dog or truffle-hunting pig," identifying potentially interesting data for human review.
• This hybrid approach allows for efficient large-scale data processing while retaining the accuracy and contextual understanding that human reporters provide.

About the Author

• Kyle Orland, Senior Gaming Editor at Ars Technica, discusses the use of AI in journalism and the balance between AI tools and human expertise.