Update unified-diffs.md

This commit is contained in:
paul-gauthier 2023-12-17 19:20:40 -08:00 committed by GitHub
parent 4ece24e8ff
commit f0cd7a6c4c
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23

View file

@ -104,8 +104,8 @@ usually intended to be consumed by the
program. program.
They need to *accurately* reflect the original and updated file contents, They need to *accurately* reflect the original and updated file contents,
otherwise the patch command will fail. otherwise the patch command will fail.
Having GPT specify changes in a well-known format that is usually consumed by a Having GPT specify changes in a format that is usually consumed by a
fairly rigid program like patch rigid program like patch
seems to encourage rigor. seems to encourage rigor.
GPT is less likely to GPT is less likely to
leave informal editing instructions in comments leave informal editing instructions in comments
@ -119,11 +119,11 @@ not talking to a person.
Aider's [previous benchmark results](https://aider.chat/docs/benchmarks.html) made Aider's [previous benchmark results](https://aider.chat/docs/benchmarks.html) made
it clear that simple editing formats it clear that simple editing formats
work much better than complex ones. work best.
Even though OpenAI provides extensive support for Even though OpenAI provides extensive support for
structured formats like json and function calls, structured formats like json and function calls,
GPT is worse at editing code if you use them. GPT is worse at editing code if you use them.
I repeated these and many other similar benchmarks against GPT-4 Turbo, I repeated these and other similar benchmarks against GPT-4 Turbo,
and again reached these same conclusions. and again reached these same conclusions.
Informally, this is probably because stuffing *source code* into JSON is complicated Informally, this is probably because stuffing *source code* into JSON is complicated