Topics2 items
AI
Items filed under this tag, newest first, across issues.
Issue 02
Can LLMs supercharge geolocations?
Bellingcat tested about ten models on 52 photos that had not been published online, and asked each for a single set of coordinates. Newer models narrowed the search much faster, and the strongest placed more than half the photos within 25 kilometres, but none was reliable for an exact point. The testers keep a person in the loop for that last check.
Why it matters
The strongest models placed more than half of 52 unpublished photos within 25 kilometres. None was reliable for an exact point.
Issue 01
Censys CLI Skills
Nine markdown skills that teach an AI coding assistant to drive cencli, the Censys command-line client for public Internet scan data. Wrappers cover search, view, aggregate, enrich, Censeye, and timeline. A CenQL reference and two methodology files tell the agent to count how common a pivot is before trusting it, and to leave the raw output next to the conclusion. Requires cencli 1.0 or later. The files are plain markdown, so they are not tied to one assistant.
Why it matters
The skills tell an assistant to count how common a pivot is before trusting it, and to keep the raw output next to the conclusion.