ChatGPT doesn't seem to have any "intelligence" about backgammon at
all and doesn't even see that its own answers make no sense.

Here is an example:
What is the 8/9/12 rule in backgammon?

The 8/9/12 rule in backgammon is a guideline that helps players determine when to run, hold, or hit in certain game positions. It is primarily applicable in backgammon races, where both players are aiming to bear off their checkers and win the race.

The rule is based on the pip count difference between the players. The pip count is the total number of pips needed to bear off all the checkers. The 8/9/12 rule suggests the following:

Run: If the pip count difference between the players is 8 pips or more, the player who is behind in the race should typically opt to run their back checkers (move them forward towards their home board) rather than engage in hitting or holding games. Running is advantageous because it maximizes the player's chances of bearing off their checkers quickly.

Hold: If the pip count difference is between 1 and 7 pips, the player who is behind should generally consider holding an anchor. Holding an anchor means keeping a checker on a higher point in the opponent's home board to create a blocking point. This strategy helps delay the opponent's progress and potentially gives the player opportunities to hit and gain an advantage.

Hit: If the pip count difference is zero or the player is ahead in the race, hitting the opponent's blots becomes a more favorable strategy. Hitting blots means landing on a single exposed checker of the opponent, sending it to the opponent's bar and delaying their progress.

It's important to note that the 8/9/12 rule is a guideline and not a strict rule. It provides general advice on how to approach certain game positions based on the pip count difference. However, other factors, such as the specific position, board structure, and risk/reward considerations, may also influence the optimal strategy in a given situation.

Players should consider the 8/9/12 rule as part of their overall understanding and analysis of the game, and adjust their strategy based on the specific circumstances they face during a backgammon race.

On 5/31/2023 9:50 AM, peps...@gmail.com wrote:
> ChatGPT doesn't seem to have any "intelligence" about backgammon at
> all and doesn't even see that its own answers make no sense.

There's no reason to expect ChatGPT to reason accurately or get its
facts right. It's a large language model, so what it's designed to
do is speak *fluently*, not do the usual things we've come to expect
from computers, such as perform correct logical calculations or act
as a search engine.

---
Tim Chow

On Wednesday, May 31, 2023 at 3:01:02 PM UTC+1, Timothy Chow wrote:
> On 5/31/2023 9:50 AM, peps...@gmail.com wrote:
> > ChatGPT doesn't seem to have any "intelligence" about backgammon at
> > all and doesn't even see that its own answers make no sense.
> There's no reason to expect ChatGPT to reason accurately or get its
> facts right. It's a large language model, so what it's designed to
> do is speak *fluently*, not do the usual things we've come to expect
> from computers, such as perform correct logical calculations or act
> as a search engine.

Understood. Ironically, the way it sounds perfectly coherent while getting
everything completely wrong, from a bg standpoint, is actually far more impressive
than if it simply gave the rule correctly. In the quote on this thread, ChatGPT is
very fluent indeed!
If it gave the 8/9/12 rule correctly, it would just be (or at least seem like) a very minor tweak on current
search-engines.

Paul

On 5/31/2023 12:36 PM, peps...@gmail.com wrote:
> Understood. Ironically, the way it sounds perfectly coherent while getting
> everything completely wrong, from a bg standpoint, is actually far more impressive
> than if it simply gave the rule correctly. In the quote on this thread, ChatGPT is
> very fluent indeed!
> If it gave the 8/9/12 rule correctly, it would just be (or at least seem like) a very minor tweak on current
> search-engines.

Exactly. Here's what I think a lot of people fail to understand. A
search-engine result is (almost always) linguistically fluent, for the
"trivial" reason that it has been *crafted by a human*, and the
engine is just regurgitating it verbatim. But a ChatGPT response is
*crafted by the AI*. Therefore, the amazing thing is how linguistically
fluent it is. It's *so* fluent that people don't even notice; they just
unconsciously treat it like a human-generated search-engine result, and
judge it by its accuracy, rather than by its fluency.

People who laugh at ChatGPT's factual inaccuracies remind me of the joke
about someone who was found playing chess with his dog. "That's
amazing!" said a passerby. "Not really," came the reply. "In our last
five games, he's only beaten me once."

---
Tim Chow