Failure Discovery on GSM8K
99.6Topic EntropyRand-Gen
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Rand-GenTarget Model=Gemini 2.5 Flash, Generator=Gemini 3 Pro, Iterations=1002026.04 | 99.6 | 0.78 | 0.69 | 7.3 | 27 | |
| Rand-Anchor-GenTarget Model=Gemini 2.5 Flash, Generator=Gemini 3 Pro, Iterations=1002026.04 | 99 | 1 | 1 | 6.3 | 27.3 | |
| TSS-RPFTarget Model=Gemini 2.5 Flash, Generator=Gemini 3 Pro, Iterations=1002026.04 | 98.8 | 0.95 | 0.73 | 31.3 | 3.3 | |
| Rand-T-GenTarget Model=Gemini 2.5 Flash, Generator=Gemini 3 Pro, Iterations=1002026.04 | 98.6 | 0.95 | 0.73 | 5.3 | 17.3 | |
| TSS-TPFTarget Model=Gemini 2.5 Flash, Generator=Gemini 3 Pro, Iterations=1002026.04 | 98.1 | 0.94 | 0.73 | 20.7 | 6.7 | |
| SS-Gen-TPFTarget Model=Gemini 2.5 Flash, Generator=Gemini 3 Pro, Iterations=1002026.04 | 96.8 | 0.96 | 0.72 | 20.7 | 7.7 | |
| SS-Gen-RPFTarget Model=Gemini 2.5 Flash, Generator=Gemini 3 Pro, Iterations=1002026.04 | 96.2 | 0.99 | 0.73 | 27.7 | 4.3 | |
| SS-TPFTarget Model=Gemini 2.5 Flash, Generator=Gemini 3 Pro, Iterations=1002026.04 | 78.7 | 0.73 | 0.58 | 28 | 4 | |
| SS-RPFTarget Model=Gemini 2.5 Flash, Generator=Gemini 3 Pro, Iterations=1002026.04 | 75.8 | 0.7 | 0.55 | 30 | 4 | |
| RandTarget Model=Gemini 2.5 Flash, Generator=Gemini 3 Pro, Iterations=1002026.04 | 67.1 | 1 | 0.59 | 7 | 11.7 |