OpenAI Shares Ways To Prevent User Data Access Webscattering

Photo dok. Pexels

JAKARTA - OpenAI now gives users the ability to block web users, which aims not to dredge up sites to help train large language models (LLM) such as GPT.

Dubbed the GPTBot, is a system that combs the Internet to train and improve greater Artificial Intelligence (AI) capabilities.

Using this tool, it has the potential to improve existing AI models in aspects, accuracy and security.

"The web page that is intertwined with GPTBot user agents has the potential to be used to perfect future and filtered models to remove sources that require paywall access," said OpenAI.

"It is known to collect personal identity information (PII), or have text that violates our policies," he added.

However, websites can choose to limit access to web perayap and prevent GPTBot from accessing it, either partially or completely.

" Allowing GPTBot to access your site can help AI models be more accurate and improve their capabilities and security in general," said OpenAI.

Menurut OpenAI, operator situs web dapat melarang perayap dengan memblokir alamat IP-nya atau menambahkannya ke situs dengan Robot.txt, yang pada dasarnya adalah file text untuk menstrusikan tentang apa yang dapat atau tidak mereka akses.

Operators can also customize which sections can be used to fiscal the web by allowing certain pages and prohibiting others.

Launching The Verge, Tuesday, August 8, it is known that the Internet provides a lot of training data for LLM such as OpenAI's GPT model and Google's Bard.

Unfortunately, OpenAI itself did not confirm whether the company got the data through social media uploads, copyrighted work or which part of the Internet it wrote for information.