OpenAI said it has found a way to get ChatGPT to avoid using em dashes, addressing a long-standing formatting pattern that has become a telltale sign of AI-generated text over the past few years.
While ChatGPT will not eliminate the em dash from its outputs by default, OpenAI said that the AI chatbot has been improved to follow user instructions to stop overusing the punctuation mark more reliably.
This means that the em dash will not entirely vanish from AI-generated text but users will now have greater control over how often it appears. ChatGPT can be made not to overuse the em dash via custom instructions in personalisation settings.
“Small-but-happy win: If you tell ChatGPT not to use em-dashes in your custom instructions, it finally does what it’s supposed to do!” OpenAI CEO Sam Altman wrote in a post on X.
In recent months, the em dash has become notorious as an indicator of AI-generated text as it started appearing everywhere from school assignments to emails, social media comments, etc. So much so that the presence of an em dash has often prompted people to label content as AI-generated even if it wasn’t.
However, many have also argued that the em dash has been a part of writing long before large language models (LLMs) adopted the punctuation.
There is no clear explanation for why LLMs tend to overuse em dashes. Since LLMs are known to output frequently seen patterns in their training data or during subsequent reinforcement learning process, online speculation suggests that it may be due to the inclusion of 19th-century books, which popularised the em dash, in models’ training datasets. Some have even pointed to automatic em-dash character conversion on the blogging site Medium as a potential reason.
While OpenAI has sought to resolve the issue, the Microsoft-backed AI startup did not offer any technical details on how it did so.
In 2023, OpenAI rolled out a feature called custom instructions that was designed to let users tailor ChatGPT to better suit their preferences. It essentially allows users to add preferences or requirements that they would like ChatGPT to consider when generating its responses.
Follow these steps to address ChatGPT’s em-dash fixation:
– Open ChatGPT
– On the web version, click on your name → Settings → Personalisation.
– Click on the Custom Instructions text field to start typing
– Type an instruction to set additional behaviour, tone, and style preferences, such as: Please do not use em dashes.
– Tap Save to save the changes.
The custom instructions option in ChatGPT is available to all, including free-tier users. However, it is unclear if the feature has been rolled out in the United Kingdom and European Union countries.
Contact to : xlf550402@gmail.com
Copyright © boyuanhulian 2020 - 2023. All Right Reserved.