Large Language Models are Zero-Shot Reasoners
Simply adding “Let’s think step by step” before each answer increases the accuracy on MultiArith from 17.7% to 78.7% and GSM8K from 10.4% to 40.7% with GPT-3.
arxiv.org/abs/2205.11916
Conversation
The truth should be simple: Text-davinci-002 (175B) or other 002 model, or instruct GPT have been finetuned with "let's think step by step. .....". I tried 001 models and none of them works with the proposed method while CoT still works well.
5
8
53
I wonder by how much performance would be worse if instead you instructed the model to start with “Obviously, the answer is”…
2
2
132
That's the trick they use to make Graduate Texts in Mathematics unnecessarily harder to understand 😂
1
96
I was literally reading this tweet. And this was the first item in my TikTok feed.
2
4
Quote Tweet
Model Psychologist is a new job type. Huge prospects in the future twitter.com/peterwildeford…
14
Might contain a lot of subtle issues, see clever Hans effect, which is always hard to debug. The law of leaky abstractions in action as my supervisor says
2
6
72
Show replies






