IC-080GPT-4's performance is largely insensitive to the content of feedback; simple re-prompting with a sound verifier (sampling) matches or exceeds detailed critique

Kaya Stechly, Karthik Valmeekam, Subbarao Kambhampati

SourceOn the self-verification limitations of large language models on reasoning and planning tasks

The paper ablates the feedback content in the iterative loop, comparing binary feedback ('wrong'), first-error feedback, and all-error feedback from a sound verifier. The differences between these conditions are small, and in two domains (graph coloring, mystery blocksworld) more detailed feedback actually decreases performance. Removing critique entirely and simply re-prompting the same question until the sound verifier accepts an answer (sampling) achieves comparable gains. A self-consistency baseline (picking the most common answer from 15 samples) shows no improvement over standard prompting, confirming that the gain comes from the sound verifier filtering, not from the model learning from feedback.

Evidence
correlational
Key metric
Table 1: Graph coloring — B.F. 38%, F.E.F. 37%, A.E.F. 34%, Sampling K=15: 40%, Sampling K=25: 44%, S.C. 14%. Blocksworld — B.F. 60%, F.E.F. 87%, A.E.F. 83%, Sampling K=15: 68%, Sampling K=25: 72%, S.C. 42%. Mystery blocksworld — B.F. 10%, F.E.F. 8%, A.E.F. 6%, Sampling K=15: 9%, Sampling K=25: 14%, S.C. 4%.
Caveat
The sampling setup has quadratically lower token cost than the critique setup because prompts do not grow with iteration count. The authors note that increasing k further can push performance higher, so the comparison at k=15/25 may understate the sampling advantage.
Model
GPT-4 / ChatGPT4 / GPT-4 Code Interpreter / GPT-4 Technical Report
Concepts
Failure mode
Datasets
Game of 24 (4nums.com instances 1-1000) [eval], PlanBench [eval]
Methods
Self-Consistency / Self-consistency prompting / Wang et al. 2023 (self-consistency) / Wang et al. 2023b (Self-Consistency) [compared-to], VAL [validation]
Related work
Tree of Thoughts [compared-to]
Related findings
IC-078, IC-079
Extraction
automatic-extraction