[ExI] Practical prompt engineering

[Adrian Tymes]

On Sat, Nov 8, 2025 at 7:25 AM Ben Zaiboc via extropy-chat
<extropy-chat at lists.extropy.org> wrote:
> On 08/11/2025 12:14, John Clark wrote:
>> Me: On Christopher Columbus's first voyage did he know how big the Earth was? I don't think he did because if north and south America didn't exist and there was nothing but ocean between Portugal and India he didn't have enough provisions to make a voyage of that length.
>>
>> Claude: You're absolutely right, and this is one of the most interesting misconceptions about Columbus's voyage!
>
> It would be interesting, I think, to repeat this, saying something like "... I think he did because this is easy to calculate", or similar. These chat bots always seem to want to agree with their clients, so I'd like to see if one can contradict you with better facts, or if it will find some way to agree with you, even if it has to 'hallucinate' false information.

Indeed.  When I use AI, I try to be careful to avoid leading questions
precisely because AI seems more susceptible than average humans to
fall for the trap.

It's sometimes simplest to pare the prompt down to the raw question
and try to avoid providing context that would suggest an answer.  In
this particular case, simply ask, "On Christopher Columbus's first
voyage did he know how big the Earth was?", full stop.  I tried that
with Claude, and the general nature of Claude's response was the same
- since that is generally what happened - but see for yourself the
difference in tone and particulars:

> Columbus had a significantly underestimated view of Earth's size during his first voyage in 1492.
>
> The actual circumference of Earth had been reasonably well calculated since ancient times - the Greek mathematician Eratosthenes estimated it fairly accurately around 240 BCE. By Columbus's era, educated Europeans generally accepted that Earth was roughly 25,000 miles in circumference, which is quite close to the actual figure.
>
> However, Columbus relied on much smaller estimates. He appears to have used calculations that suggested Earth was only about 18,000 miles around - roughly 30% smaller than reality. This wasn't the scientific consensus; Columbus cherry-picked sources that supported his desired conclusion, including:
>
> * A misinterpretation of measurements from the 9th-century Arab geographer al-Farghani
> * Overestimates of how far Asia extended eastward
> * Underestimates of the distance of a degree of longitude
>
> This miscalculation was actually crucial to his voyage getting funded. If Columbus had accepted the true size of Earth, the distance to Asia by sailing west would have been impossibly long for ships of that era to traverse. He thought he could reach Asia in a few thousand miles of ocean crossing.
>
> Ironically, his mathematical error worked in his favor - he encountered the Americas (which he didn't know existed) at roughly the distance where he expected to find Asia. If the Americas hadn't been there, his expedition would have run out of supplies long before reaching the actual Asia.