Home | Notifications | New Note | Local | Federated | Search | Logout

Note Detail

Reply to @7666@comp.lain.la
Q.U.I.N.N.@icedquinn@blob.cat (2026-09-29 10:57:59)
@7666 @sun i shipped one project on glm 4.7 and probably could have finished the second one with it. i just realized that it was enabling me to write code badly and stopped.
---Reply---
Blurry Moon@sun@shitposter.world (2026-09-29 10:59:49)
@icedquinn @7666 so like another thing is: if you have super good project planning discipline, test-first, etc, then it makes up for the deficiencies a LOT.

---Replies---

Q.U.I.N.N.@icedquinn@blob.cat (2026-09-29 11:03:54)
@sun @7666 a big problem i have is that model based programming is just a better paradigm in the first place.

ex. with GSL you run scripts against xml to generate the code. in the parsing machine i was trying to rush through a lot of it is regular. there's opcodes, there's behaviors to interpret them, all the IDs need to line up, there's a document. putting it in a model file and deriving the artifacts literally deletes classes of mistakes entirely.

GLM slopping is "i updated the grammar sheet, now i have to ask the machine to fix it, then i have to verify that it did," and the leverage from this is very fucky.

with refal or maude your artifacts come out of a guaranteed pipeline from the model. i've been trying to read the maude manual to try this out before evangelizing it too hard.

7666@7666@comp.lain.la (2026-09-29 11:07:40)
@sun @icedquinn this is probably why i don't need significantly stronger models. i do a lot of the planning legwork ahead of time. once i actually have a vision of what i want and how to get there my prompt basically lays it all out with my success criteria and all.