AI chatbots and LLM’s are an amazingly impressive development that will have a great impact on OSINT and intelligence analysis. Many of you use chatbots may be even on a daily basis for a variety of tasks and assignments, from simple ones to more complex things. And results can be surprising.
That is, surprisingly good but also surprisingly bad. ChatGPT, Gemini and Copilot alike, sometimes simply do not do what they were asked to do, leave out things, add things I did not ask for, or make errors. The first two can be solved by asking very specific questions and make clear exactly what you want. Explicitly mentions sources to be used and the required output can help a lot.
Unfortunately, the bots return many errors in their answers. Sometimes so obvious that it becomes a little annoying. Especially when asking for factual data, or solutions to problems, the bots in my experience return many errors. I have therefore developed the habit of always do the following.
The second prompt I enter is by definition: correct the errors.
Regardless what the first answer was, the second prompt is always the same: ‘correct the errors’. In all cases so far, the bot apologises, admits the errors and mistakes in the first answer, explains that it should have done a better job, promises it will not happen again (blabla) and produces a new answer. The result is really remarkable.
I call this procedure: The 2nd Prompt Principle.
As a habit, always apply this principle with more complex questions. Also for the 3rd and maybe 4th prompt. Very curious how long this will last. For now, I am still very cautious in using AI generated answers. I am sure I am not the only one.
