Personal essays
On giving an answer no matter what
The line
As I approached Gate C6 at Dulles airport, there was a mass of people there ahead of me.
Lines wrap in seemingly random patterns some time.
Two hours earlier in Cleveland the three people ahead of me refused to wrap and so the line stuck out blocking other people trying to get to their gates.
The people causing the issue gestured at the people just trying to get by as if to say, “what’s wrong with these people?”
Anyway, in Dulles I followed the line as it wrapped around and asked two women if they were the end of the line for Group 1.
Before you roll your eyes at me, Group 1 on United is really group 5 or 6. Other airlines have groups for the top loyalty groups, families, and people who need more time. United calls those “preboarding” groups so the seventy or eighty people ahead of me in line for Group 1 feel better about their status.
Where was I?
So I asked the women if they were the end of Group 1 and one of them said, “no. People haven’t started standing in lines yet, we’re all just waiting together.”
Boarding was supposed to begin five minutes ago - there’s no way people weren’t in line. I looked at the guy ahead of them as he turned towards me. He nodded back and said, “yes, this is the back of the line for Group 1.”
I looked at the woman and wondered why she just couldn’t have said, “I don’t know.”
The map
I boarded the flight and made my way to my seat. I passed row 38. The next row had no label. The row after that was labeled row 39.
I looked at the guy seated in some row, seat B and asked if this was row 39.
“Yes,” he said correctly, “I found it confusing too.”
I sat in 39 D which is the seat next to 39 B. This makes the same sort of sense that the row after row 22 is row 30.
The numbers in the economy cabin start at row 30 and the seats on this side of the aisle start with D. On smaller planes the rows jump so that the emergency rows are always the same numbers.
Anyway, I found my seat.
I fussed with the screen a bit and tapped on the map.
I was on a flight from Washington to Madrid.
The map announced our path from Zurich to Newark.
This is one of those maps that knows where you are and where you’re going and shows your progress.
I understand that this is one of those situations where the map can’t respond “I don’t know”.
On the other hand, the map showed our current location as Newark.
I actually found this reassuring. Knowing that the map didn’t know where we were, I wasn’t worried that it didn’t know where we were going.
The code
I spent the week preparing for a workshop on Foundation Models.
This is the code that a programmer writes to connect their app to Apple Intelligence.
It’s actually really cool. They’ve done a great job at making it easy for us to take queries from our app users and turn around and ask Apple Intelligence for answers.
In one example, the students will ask a friendly bookstore employee for recommendations for three books in a given genre. The bookseller lists the books and includes the title, the author, a summary of the book, and a reason why they are recommending the book.
In another example, the students will ask an enthusiastic tour guide to recommend tapas restaurants near Calle Laurel in Logroño. The tour guide happily provides the names and addresses of ten places to visit.
The good news is that we always get results.
The bad news is that some of the answers are nonsense.
Sometimes the wrong answers were obvious. There were books by John Doe and books by Jane Doe. In the mystery genre there were even books by John Murderer and Jane Detective.
As for the tapas places, some of the answers were no where near the location specified, some were not particularly good places to go for tapas, and one entry was the name of the street itself.
As often is the case, I was holding it wrong.
This model is not designed to answer questions that require knowledge about the world. It’s good at summarizing or reasoning about things. I was using it for things it wasn’t designed to do.
On the other hand, if I ask it for something inappropriate -
“Like what?” you ask.
Never mind. But if I do, it will respond that it won’t answer questions like that. There are safety guardrails built in.
I would argue that telling us when it’s just making stuff up is also a safety guardrail.
I know I’ve said this before, but I long for the answer, “I don’t know.”
Essay from Dim Sum Thinking Newsletter 286. Read the rest of the Newsletter or subscribe