A sharp critique says LLMs are speeding up forgery, not trustworthy coding

teluyiyu · reddit · 2026-07-23

This discussion highlights a long essay arguing that LLMs are better understood as tools for forgery than for hallucination: they let people produce convincing output faster, but without reliable provenance.

The essay’s main points are:

junior devs using vibe coding can overproduce polished PRs that still deserve skepticism;
open-source maintainers are seeing more slop PRs from people chasing GitHub credit;
“10x output” is not automatically a win when every line of code becomes liability;
software culture has accepted AI-generated output more readily than games or other creative fields;
the proposed fix is not an outright ban, but mandatory source attribution alongside inference, until models can prove provenance.

The Reddit post also raises the copyright question: if no court should have ruled on AI output’s copyrightability because the outputs are unsourced, that itself would be a major point of contention.

Original post →

A sharp critique says LLMs are speeding up forgery, not trustworthy coding

More from coding & agent