Stata Translation Workshop · Post-Workshop Survey
Drawn only from "What concerns (if any) do you have with using large language models for code translation?" (all 9 respondents answered). Same style as the earlier clouds: packed into a circle, right angles only, tight regular spacing. Size and color track how many respondents raised each concern; generic subject-matter words (code, Stata, Claude…) stay small and gray.
Two concerns tie for the top spot (2 of 9 respondents each): fearing the translated code is subtly wrong and hard to catch, and needing better ways to test/control the model's output. Every other concern was raised by exactly one respondent — overconfidence in incorrect code, silently simplified code going unflagged, checking accuracy on large packages, code that isn't idiomatic to the target language, maintaining code nobody wrote by hand, model stochasticity, who's accountable for the output, trusting it for real research decisions, unfamiliar shell/PATH setup, and translating new features incrementally rather than redoing everything.
| Concern | Respondents | Original text |
|---|---|---|
| Fear of hidden errors | 2 | "the code may be wrong and it would take a lot of time to find the errors" · "worry that translated code might be wrong" |
| Testing & output control | 2 | "a way to control what the model spits out, by creating very good testing files" · "how to test the translated code" |
| Overconfidence / false correctness | 1 | "Code that appears to be correct but is not - over confidence." |
| Silent simplification | 1 | "My code got simplified by Claude and it just put a comment… # simplified" |
| Flagging changes | 1 | "a skill saying to Claude to report back if it does any simplifications or approximations" |
| Manual verification effort | 1 | "check it line-by-line and do extensive testing" |
| Accuracy at scale | 1 | "How do we check accuracy, especially for large complex packages" |
| Maintainability | 1 | "How can packages be updated/maintained/checked if not human-written?" |
| Code elegance & idiomaticity | 1 | "Lack of elegance (e.g. using optimal structures/organisation/logic for each coding language)" |
| Stochasticity & reproducibility | 1 | "The intrinsic stochasticity…" |
| Accountability | 1 | "the question of who is responsible for the code" |
| High-stakes trust | 1 | "particularly if you're trusting it for scientific research and decision making" |
| Shell & script unfamiliarity | 1 | "Lots of requests are shell script code which I didn't quite understand" |
| Environment setup | 1 | "Need to make sure to set up PATH variables correctly" |
| Incremental re-translation | 1 | "find a way to translate… new feature… without re-translating everything" |