I had the opportunity to use AI to speed up my work yesterday - and I didn't take it.
I am updating some old lecture notes, and needed to replace an obolete typesetting command with a new command. Essentially a search & replace job, but the syntax of the two commands is different enough that the editor's textual search & replace wouldn't work. I used the inbuilt search, but then had to change each occurrence manually.
Or I could have uploaded the file into Copilot and asked it to make all those changes for me. Why didn't I?
Because then my entire module's lecture notes would have passed through a highly complex process I do not understand or trust. I would not feel confident that the AI had only changed the things I asked it to change.
When my notes contain human mistakes - hardly a rare occurrence - I feel fine apologising to the students and making the corrections.
But if my notes contained weird, unexplained, AI-generated errors, I expect I'd feel angry & embarrassed.
I suppose could ask a second AI to compare the output against the original - I didn't actually think of this at the time. I suppose that would provide some reassurance, but would it be enough? Is that the answer to these conundrums?
Richard Elwes
npub1m4…kf85f
2026-01-15 10:32:22 UTC
Author Public Key
npub1m4zfe40awg0pgu40l7gy4qxcwgkh9t97pwa7csuefv5ps60dmv2slkf85fSeen on
wss://relay.momostr.pinkPublished at
2026-01-15 10:32:22 UTCEvent JSON
{
"id": "13f5e80faa2e7e4aac04a4bb51afd84062a9e34d84501c3f5291ee5f75c0bb07",
"pubkey": "dd449cd5fd721e1472afff904a80d8722d72acbe0bbbec43994b281869eddb15",
"created_at": 1768473142,
"kind": 1,
"tags": [
[
"proxy",
"https://mathstodon.xyz/@richardelwes/115898655868873202",
"web"
],
[
"proxy",
"https://mathstodon.xyz/users/richardelwes/statuses/115898655868873202",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://mathstodon.xyz/users/richardelwes/statuses/115898655868873202",
"pink.momostr"
],
[
"-"
]
],
"content": "I had the opportunity to use AI to speed up my work yesterday - and I didn't take it.\n\nI am updating some old lecture notes, and needed to replace an obolete typesetting command with a new command. Essentially a search \u0026 replace job, but the syntax of the two commands is different enough that the editor's textual search \u0026 replace wouldn't work. I used the inbuilt search, but then had to change each occurrence manually.\n\nOr I could have uploaded the file into Copilot and asked it to make all those changes for me. Why didn't I?\n\nBecause then my entire module's lecture notes would have passed through a highly complex process I do not understand or trust. I would not feel confident that the AI had only changed the things I asked it to change.\n\nWhen my notes contain human mistakes - hardly a rare occurrence - I feel fine apologising to the students and making the corrections.\n\nBut if my notes contained weird, unexplained, AI-generated errors, I expect I'd feel angry \u0026 embarrassed.\n\nI suppose could ask a second AI to compare the output against the original - I didn't actually think of this at the time. I suppose that would provide some reassurance, but would it be enough? Is that the answer to these conundrums?",
"sig": "599c8ef7ce6d19da6a2dd94b9aa82d41bf63b0a7c629f6e25d77bf6d6325d924c6d018cee621e33217f1f6a7b86b2c6e8f4677b86ada8be9814c55c2fa07509e"
}