Rewrite Text: ChatGPT vs. text-davinci-003

I’m trying to rewrite texts in a more formal/academic way. ChatGPT does a fantastic job but text-davinci-003 just doesn’t get it right - the texts are either too short or almost identical to the original. I played a bit with the temperature parameter but it doesn’t help much. Max_tokens=1000, stop=None

My latest prompt: f’Rewrite the following paragraph in an academic style for a journal in the human-computer-interaction field. Do not make it shorter.\n\n{paragraph}’

For example:

“The main part of the experiment was where the subjects made decisions that affected their payoff and interacted with a simulated AI. To simulate a decision and advice situation, we used the Judge-Advisor-System (JAS). Here, a subject in the role of a decision maker had to decide on an estimation. In the first step, the subject made the first estimation and then received an advice from the advisor (here the simulated AI). After that, the subject submited the second estimaiton. Thus, the participant made two estimation. The extent to which the subject accepted the advisor’s advice for the second estimation could be measured for variables with a metric scale level by Weight of Advice (WOA). WOA was the independent variable in our experiments: the higher the WOA, the more the subject accepted the advice of the AI.”

“In the main portion of the experiment, the participants were tasked with making decisions that impacted their payoff while interacting with a simulated artificial intelligence (AI). In order to simulate a decision-making and advice scenario, we employed the Judge-Advisor-System (JAS). During this process, the participants, serving as decision makers, were required to provide an initial estimation. They then received advice from the simulated AI, after which they submitted a second estimation. This resulted in a total of two estimations being made by the participant. The extent to which the participant accepted the AI’s advice for their second estimation was quantified using the Weight of Advice (WOA) metric. WOA served as the independent variable in our experiments, with a higher WOA indicating a greater acceptance of the AI’s advice.”

“In our experiment, we utilized the Judge-Advisor-System (JAS) to simulate a decision and advice situation. Subjects were placed in the role of decision makers and were required to make two estimations. The first estimation was made by the subject, followed by an advice from the simulated AI. The second estimation was then submitted by the subject. The extent to which the subject accepted the advice of the AI was measured using the Weight of Advice (WOA) metric scale, with higher WOA indicating greater acceptance of the AI’s advice. WOA was the independent variable in our experiments.”

1 Like

Telling GPT NOT to do something doesn’t always work as well as telling it what you WANT it to do

Instead of saying do not make it shorter, you could say be verbose and make sure every facet of the original content is covered.

Good luck


Thanks for the quick reply. It doesn’t help much but I will remember it for the future.

Any other ideas?

Try using “lets think step by step” in your prompt. It did help me a lot to summarize a transcription the best way possible. I am still looking for better ways to summarize transcriptions but its the prompt that have better worked for me.

1 Like