SOMETHING WONDERFUL HAS HAPPENED TO YOUR leah :oh_no: on Nostr: funny thing. in contrast to the veritable *gush* of people who have suddenly adopted ...
funny thing. in contrast to the veritable *gush* of people who have suddenly adopted vibecoding and even though you'd think they'd know better they're basically all saying the same things about how great it is...
i can't help but think back to Chuck Moore's recent - and by his own description, miserable - experience of trying to get Microsoft's own LLM to figure out how to use the Microsoft API that stopped working the way colorForth expected it to.
one small, simple problem - and he couldn't cajole the LLM from the company that made it to help him solve it.
so i guess the lesson to derive is that LLMs are great if all you care about is spewing out code that passes the test suite. but actually *debugging* that code? not gonna happen. the LLMs can't do that.
*because stochastic parrots aren't capable of analysis*.
Published at
2026-03-05 16:11:05 UTCEvent JSON
{
"id": "33bc7ab9eb786a2224a6077734725bd3be252b007f5160165a16aaeb8877e053",
"pubkey": "aa6fb3a41adf045c1169359a7745dfa055b8dec7f9bd7b6dc5160e7ee371a575",
"created_at": 1772727065,
"kind": 1,
"tags": [
[
"proxy",
"https://oldbytes.space/@millihertz/116177440973864797",
"web"
],
[
"proxy",
"https://oldbytes.space/users/millihertz/statuses/116177440973864797",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://oldbytes.space/users/millihertz/statuses/116177440973864797",
"pink.momostr"
],
[
"-"
]
],
"content": "funny thing. in contrast to the veritable *gush* of people who have suddenly adopted vibecoding and even though you'd think they'd know better they're basically all saying the same things about how great it is...\n\ni can't help but think back to Chuck Moore's recent - and by his own description, miserable - experience of trying to get Microsoft's own LLM to figure out how to use the Microsoft API that stopped working the way colorForth expected it to.\n\none small, simple problem - and he couldn't cajole the LLM from the company that made it to help him solve it.\n\nso i guess the lesson to derive is that LLMs are great if all you care about is spewing out code that passes the test suite. but actually *debugging* that code? not gonna happen. the LLMs can't do that.\n\n*because stochastic parrots aren't capable of analysis*.",
"sig": "c364ca6b6b80a8a15e22f7a84395868c20b32f66cff97ea34fef4894e6314290b94c4278b6800a02ae1f21fbb2dbfece75ffda2a4377df342c96af5ac0e663a4"
}