4.153 blood, laundry, a/v, bots

Wednesday 05/03/2023

First thing I was to head out for an 8:30 appointment for a routine blood draw ordered by my PCP. But it is also laundry day, my slot for our 6th floor machines is 8-12. So I started a load at 7 and was able to start the second load just before I left. Down to Sunnyvale to the clinic — for some reason, the PAMF website didn’t even offer blood draw appointments at the nearby clinic. NPO of course.

Then, hungry for breakfast, I stopped at the cafe in Midtown and had an omelet. So back to CH near 10 to rotate the laundry.

Yesterday during the event on the 11th floor, the new microphones had seemed to have a problem where they would cut out for a second or two, intermittently. As I noted, that seemed to mean I should return them. But I wanted to reproduce the problem. So now I went up there and set up the system. I was also listening to a podcast on the phone, so I set the phone on a table playing the podcast, and laid a mic on the phone so I could listen to the amplified podcast coming out of the speaker. For 20 minutes. And it didn’t cut out, it sounded fine.

Now what? I dunno. I got to talk to Bert.

Later in the day I got my official invite to try Google Bard. It gave me some reasonable responses to health queries and general knowledge. Then I decided to try it on code. Something I know about is the use of the Qt GUI platform with Python, PyQt. So I asked it to write me a specific piece of code using Python 3.10 and PyQt6.

It kind of did, reeling off 100 lines of some credible looking code, but when I pasted that into my development system and tried to run it, I quickly found it had made several mistakes. I went back and told it about those errors. It apologized profusely and gave me some revised code,. That also had errors, different ones.

A friend here who I told about this, asked if chatGPT4 could do any better. So I tried it. I will say that GPT4 gave me what I would call better-structured code, better-organized and more readable, but it also made the same kinds of errors.

Ironically the bots are being tripped up by the same thing human developers — well, me at least — have been having problems with. When Qt went from version 5 to version 6 they changed a number of naming conventions, so that code that worked in Qt5 had to be edited to do things like change the name Qt.KeyFind to read Qt.QKeySequence.KeyFind. Many, many such changes. It appears both bots have been “trained” on probably gigabytes of PyQt5 code and even though I instructed them specifically to use PyQt6.5, the code they regurgitate mixes in some of the old style names.

The problem is, you don’t know they’ve done it until Python tries to execute a statement with an incorrect name. Only then — because Python is not a compiled language where everything is checked at compile time — you get a run-time error. Which means one of these can lurk in what seems to be working code until the right combination of inputs sends it to the line with the error. And then you have an annoying and time-consuming search through the documentation trying to find the name that can’t be found, and figuring out how it should be rewritten for the current version.

Early on some users created automated scripts that would rip through an old program and fix all the names to the new style. But that should not be necessary with code supposedly written to the current standard. Which is what the bots claimed to be doing.

Anyway for this particular little corner of the coding world, I demonstrated that neither bot can be trusted yet.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s