Agentic don't tend to read and/or follow README.md

Neither o3 nor claude-3.5-sonnet tend to read README.md in agentic mode (like “males never read instructions”).

However claude-3.5-sonnet is a bit smarter on improving on this mistake.

In example below README.md tells that Rust code should be targeting the native platform (execution examples are given for build and run, but not for tests) therefore tests are assumed to be executed like RUSTFLAGS="-C target-cpu=native" cargo test ...

Update: even if you explicitly put README.md into the context the result is the same.

Here is o3-mini:

Here is claude-3.5-sonnet:

its starting to behave too much like a human, we might need it to rollback :joy:

2 Likes