Post #4113968
2026-07-26 14:27 UTC
@levitte@mastodon.nu @rkaj@mastodon.nu I would need to have a scenario explained to me where an LLM is used to generate code and a subsequent review where the person walks away with as much understanding of the code as if the person had written the code themselves.
Replies (2)
-
@ahltorp@mastodon.nu 2026-07-26 14:30
@levitte@mastodon.nu @rkaj@mastodon.nu Which, by the way, was part of the criteria we used when I supervised programming students in 1996.
-
@levitte@mastodon.nu 2026-07-26 17:00
@ahltorp@mastodon.nu @rkaj@mastodon.nu Not sure what you're looking for in terms of scenario, but perhaps a summarised session could do? 0. Human chooses a task to be done, and describes it to the agent, with criteria and guardrails. 1. Agent produces something. 2. Human reads what agent has produced. 3. If human understands and likes, the product is shipped (committed, packaged, whatever), and the task is done 4. Otherwise, human asks for explanations, or opines, and gives the agents refined instructions. 5. goto 1