Wednesday, January 18, 2023


 An interesting new search engine experiment in which you put in a question, it searches a very large quantity of books (in the thousands, I believe), and ChatGPT writes a summary of it. 

The limitations of its current incarnation can be seen with the answer it gave me to "What is a Fesapo syllogism?", with Speed rather than Quality selected:

A Fesapo syllogism is a syllogism with a major premise, a minor premise, and a conclusion all using propositions of the form A, where the major premise is Cesare, the minor premise is Camestres, and the conclusion is Festino.

This is total gibberish, of course. The vowels of names of syllogisms give you the forms of the major, minor, and conclusion, so the propositions of Fesapo are E, A, and O respectively. (The only syllogism in which major, minor, and conclusion are all A is Barbara. When I asked it what a Barbara syllogism was, it said that it was a syllogism with a major premise, a minor premise, and a conclusion [true but uninformative] and gave a correct example.) Cesare, Camestres, and Festino are names of other syllogisms, not names that apply to propositions.

When I searched with Quality rather than Speed, it gave me an answer with less gibberish that was nonetheless still gibberish:

A Fesapo syllogism is a valid syllogism made up of major premise A, a minor premise, and a conclusion all using propositions of the form A.

One of the nice things about the search is that it gives you the texts that primarily influenced its summary; in this case, it is largely influenced by works like Clifford A. Pickover's The Loom of God, which nonetheless is talking about Barbara. Barbara is of course far and away the most commonly used syllogism name, and Fesapo is a name very rarely used and rarely if ever used as an example, so almost inevitably any text that mentions the name 'Fesapo' will also have nearby an explanation of Barbara.

On the other side, it could not come up with anything for Festino (despite the fact that it's pretty much guaranteed to be mentioned in any context in which Fesapo is).

This is its answer for "Who is Jesus?" (Quality selected):

Jesus is all-powerful (Isaiah 9:6), has authority over demons (Mark 1:27), is the Son of God (Luke 1:35), is God (John 1:1-5), is the Messiah (John 4:25-26), is the Judge (John 5:22), gives life (John 10:10), is the Good Shepherd (John 10:11), is the only way to God (John 14:6), is the author of life (Acts 3:15), is the wisdom of God (1 Corinthians 1:21-24), is the head of the church (Ephesians 5:23), is the savior of humanity (Matthew 1:21), and is the anointed one who fulfills the promise of a coming Messiah from the line of David who will deliver the Children of Israel and usher in a new age of peace and God’s rule (Don't Know Much About® the Bible: Everything You Need to Know About the Good Book but Never Learned)

Despite the bit at the end, almost all of this seems to be lifted from The One Year Chronological Bible.

Its explanation of Aristotle's account of virtue was OK, although incomplete, and seems to have been drawn heavily from books by Bernard Williams. When specifically asked to give Aristotle's definition of virtue, however, it gave a correct answer. It also correctly identifies the point in dispute between rationalists and empiricists, gives an OK although not completely accurate summary of Anselm's ontological argument (which it takes from Daniel Dennett), and gives a correct summary of the distinction between a valid argument and a sound argument.

It gave a basically correct summary of four of the Five Ways of Thomas Aquinas, but this is its account of the Third Way:

The Third Way of Thomas Aquinas is the intuition of things previously known only by revelation.

This is what it gave me when I asked for a recipe for fudge:

2¼ cups sugar 2⁄3 cup pure maple syrup 2⁄3 cup heavy cream 2 tablespoons butter 1 teaspoon vanilla extract In a medium saucepan, combine the sugar, maple syrup, cream, and butter. Cook over medium heat, stirring constantly, until the mixture reaches 234°F on a candy thermometer. Remove the pan from the heat and stir in the vanilla. Pour the mixture into a buttered 8-inch square pan. Let cool until just warm, about 10 minutes. Using a wooden spoon or a heatproof rubber spatula, beat the fudge until it thickens and loses its gloss, about 5 minutes. Pour into the prepared pan and let cool completely. Cut into 1½-inch squares.

This is of course a recipe for maple fudge, not ordinary fudge; the basic structure is correct but, if you will pardon the pun, the numbers are fudged: the numbers for sugar and syrup seem to be an odd proportion, 234 degrees seems reasonably close to what you'd want, but in a real recipe you'd always find it rounded up or down. I'd be surprised if beating the fudge for only five minutes would have much effect. Nonetheless, without actually trying it out, this is probably quite edible, although it might end up being closer to caramel than fudge. I couldn't tell from the results it gave what its major source was.

It correctly identifies the main character of The Hobbit as Bilbo Baggins. Its answer to the question "Who is Sauron?" is also correct:

Sauron is a manifestation of the power of darkness upon earth, and is also known as Zigūr the Great. He is an obsessed being, driven by his fever to dominate everything and everybody. He is the servant of Morgoth, and is self-corrupted.

There's a bit of unusual trivia for you; Zigur is the Numenorean name given to Sauron in Tolkien's The Notion Club Papers, 'zigur' being the Adunaic word for 'wizard'. Its answer to "Why did Galadriel reject the One Ring?" is also correct:

Galadriel rejected the One Ring because she knew that if she accepted it, she would be tempted by its power and potentially become corrupted by it. She humbly chose to reject it in order to remain true to her mission of helping the free peoples in their battle against evil.

 It however gets the Elvish word for 'friend' wrong (famously mellon, although of course Tolkien has others). Its answer, though, was "Elendili"; which means 'Elf-friends'.

It's an interesting way of looking at how these kinds of large language model algorithms work.