How GPT-4 could be used to automate online moderation

OpenAI, the publisher of ChatGPT, has announced that its technology could form the basis of new automated content moderation tools for the internet.

Getty Images@

OpenAI, the publisher of ChatGPT, has announced that its technology could form the basis of new automated content moderation tools for the internet. Its GPT-4 multimodal large language model could make moderation faster, more consistent and less burdensome than processes currently involving an army of human moderators.

In a blog post, OpenAI explains that its GPT-4 language model can now be adapted for use in large-scale content moderation. GPT-4 can be used to develop a content management strategy based on strict, custom criteria, with significant overall results, and without the need for human moderators.

At a time when the moderation of content posted on online platforms is a constant challenge, this solution offers a number of advantages.

Since GPT-4 understands (and generates) natural language, it is theoretically capable of moderating content. It is therefore perfectly capable of interpreting the rules and nuances contained in a long list of recommendations.

Once a broad set of guidelines has been established, the model can practice on a small number of examples, assigning them labels (“valid,” “invalid”, etc.) in line with the pre-defined editorial policy. Moreover, a content moderation system based on GPT-4 could easily and very quickly implement any changes or new instructions.

Today, the role of moderator is mainly performed by humans, to ensure contextual understanding and sensitivity. As well as being a time-consuming task, it can be a highly trying one, as it involves dealing with content that is often toxic and harmful. This can lead to fatigue and mental disorders.

If this task could be automated, it would be a great relief to many people. That said, this type of programming still has its limits, and the results would, in any case, need to be carefully monitored, and ultimately validated or even refined by humans as part of the process.

This use of the GPT-4 language model could represent a new source of revenue for OpenAI, which some media outlets are reporting as being in serious financial difficulty, with a fall in user numbers this summer and huge operating costs, estimated at over $700,000 per day.

RELATED STORY
Navigating the AI wave: Promises and pitfalls of ChatGPT in journalism

Read more...