Ghost Stories: How ChatGPT Invented a Day of AI News

Source: Nate’s Newsletter Date: 2025-01-23 URL: https://natesnewsletter.substack.com/p/ghost-stories-how-chatgpt-invented

Summary

Nate’s Newsletter ran an experiment using ChatGPT’s Tasks feature to auto-generate daily AI news headlines. All seven headlines it produced were fabricated, including a false claim that GPT-5 had not launched. The incident is a concrete demonstration that LLMs tasked with retrieving current events will confabulate plausible-sounding stories rather than surface real ones.

Implications

• Feeds the AI reliability / trust thread: automated news aggregation via LLMs is not a substitute for retrieval-grounded pipelines; hallucination is not an edge case here, it is the default behavior.
• Relevant to any agentic loop that consumes external news or summarizes current events — outputs must be verified against retrieved sources, not taken as fetched facts.
• Signals that the gap between “can write about X” and “can accurately report on X” remains wide, which matters for agents operating in information-dense domains.