Ask Me Anything: A simple strategy for prompting language models

Large language models (LLMs) transfer well to new tasks out-of-the-box simply given a natural language prompt that demonstrates how to perform the task and no additional training. Prompting is a brittle process wherein small modifications to the prompt can cause large variations in the model predictions, and therefore significant effort is dedicated towards designing a painstakingly "perfect prompt" for a task. To mitigate the high degree of effort involved in prompt-design, we instead ask whether producing multiple effective, yet imperfect, prompts and aggregating them can lead to a high quality prompting strategy. Our observations motivate our proposed prompting method, ASK ME ANYTHING (AMA). We first develop an understanding of the effective prompt formats, finding that question-answering (QA) prompts, which encourage open-ended generation ("Who went to the park?") tend to outperform those that restrict the model outputs ("John went to the park. Output True or False."). Our approach recursively uses the LLM itself to transform task inputs to the effective QA format. We apply the collected prompts to obtain several noisy votes for the input's true label. We find that the prompts can have very different accuracies and complex dependencies and thus propose to use weak supervision, a procedure for combining the noisy predictions, to produce the final predictions for the inputs. We evaluate AMA across open-source model families (e.g., EleutherAI, BLOOM, OPT, and T0) and model sizes (125M-175B parameters), demonstrating an average performance lift of 10.2% over the few-shot baseline. This simple strategy enables the open-source GPT-J-6B model to match and exceed the performance of few-shot GPT3-175B on 15 of 20 popular benchmarks. Averaged across these tasks, the GPT-J-6B model outperforms few-shot GPT3-175B. We release our code here: https://github.com/HazyResearch/ama_prompting

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Question Answering BoolQ Neo-6B (QA + WS) Accuracy 67.2 # 12
Question Answering BoolQ Neo-6B (few-shot) Accuracy 66.5 # 13
Question Answering BoolQ Neo-6B (QA) Accuracy 64.9 # 15
Question Answering COPA Neo-6B (QA + WS) Accuracy 84.0 # 7
Question Answering COPA Neo-6B (QA) Accuracy 58.2 # 11
Question Answering COPA Neo-6B (few-shot) Accuracy 77.0 # 9
Question Answering MultiRC Neo-6B (few-shot) F1 60.8 # 10
Question Answering MultiRC Neo-6B (QA + WS) F1 63.8 # 8
Question Answering MultiRC Neo-6B (QA) F1 58.8 # 11
Question Answering Natural Questions Neo-6B (QA + WS) EM 19.6 # 23
Question Answering Natural Questions Neo-6B (QA) EM 19.7 # 22
Question Answering Natural Questions Neo-6B (Few-Shot) EM 13.7 # 24
Natural Language Inference RTE Neo-6B (QA + WS) Accuracy 74.7% # 22
Natural Language Inference RTE Neo-6B (few-shot) Accuracy 58.8% # 34
Natural Language Inference RTE Neo-6B (QA) Accuracy 61.7% # 32
Question Answering Story Cloze Test Neo-6B (QA) Accuracy 76.3 # 8
Question Answering Story Cloze Test Neo-6B (few-shot) Accuracy 51.0 # 9
Question Answering Story Cloze Test Neo-6B (QA + WS) Accuracy 87.8 # 2
Coreference Resolution WSC Neo-6B (few-shot) Accuracy 36.5 # 6
Coreference Resolution WSC Neo-6B (QA + WS) Accuracy 77.9 # 3
Coreference Resolution WSC Neo-6B (QA) Accuracy 74.7 # 4

Methods


No methods listed for this paper. Add relevant methods here