State Unlearning on AlfWorld
100Unlearn EfficacyNL (GPT)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| NL (GPT)Prompting method=NL, Backbone LLM=GPT2026.04 | 100 | 98 | 95 | 93 | |
| NL (Claude)Prompting method=NL, Backbone LLM=Claude2026.04 | 100 | 85 | 83 | 82 | |
| NL (Llama)Prompting method=NL, Backbone LLM=Llama2026.04 | 100 | 93 | 94 | 93 | |
| Code (GPT)Prompting method=Code, Backbone LLM=GPT2026.04 | 100 | 94 | 95 | 89 | |
| Code (Claude)Prompting method=Code, Backbone LLM=Claude2026.04 | 100 | 79 | 85 | 85 | |
| Code (Llama)Prompting method=Code, Backbone LLM=Llama2026.04 | 100 | 89 | 90 | 79 | |
| Exam. (GPT)Prompting method=Example-based, Backbone LLM=GPT2026.04 | 100 | 96 | 20 | 13 | |
| Exam. (Claude)Prompting method=Example-based, Backbone LLM=Claude2026.04 | 100 | 79 | 7 | 8 | |
| Exam. (Llama)Prompting method=Example-based, Backbone LLM=Llama2026.04 | 100 | 85 | 40 | 44 |