If you’re not familiar with the GPT-2 algorithm of OpenAI, let me ask you this, “Have you ever found yourself having difficulty writing a handful of sentences and paragraphs, and then you received an unsatisfactory school remark?

You might want to look for a miracle in doing things effortlessly and hope that your computer would get your point and do the work better for you. It sounds impossible, but that was the case with a grad student named Tiago.

What got Tiago seamlessly through his requirement as a graduate student in business school was an algorithm by OpenAI, called GPT-2. The GPT-2 is a system that works on language modeling, where you predict cohesive paragraphs after analyzing texts from given sentences and quotations. Tiago fed the algorithm with sentences of his thoughts, and the robust tool continued each paragraph.

This new technology seems interesting, but OpenAI worried that the algorithm would be misused to form malicious spam, fake news, and propagandas with these possibilities. If you feed it with fake topics, it would surely finish it for you.

So, in their fear that releasing the full version would be too dangerous, OpenAI released a smaller model in February 2020. Adam King’s Talktotransformer.com can publicly access this model.

Advantages and Disadvantages of Using This Artificial Intelligence

The use of this technology indeed has its own set of pros and cons, as follows:

Advantages

  • The algorithm is flexible.
  • Its 1.5 billion parameters were trained on a dataset comprising 8 million web pages, which generates texts from studying patterns from these pieces of information.
  • It notices a variety of inputs and also identifies infamous characters.
  • It can function as a dungeon master as well, where it steers players through a text adventure based on the player’s commands.
  • You can write it along with literature, news article, stories, and HTML.

Disadvantages

  • It can be more misused if it becomes more reliable and consistent.
  • The system has misleading tendencies.
  • The algorithm does not fully grasp the matter of language or the world itself.
  • The algorithm can be incoherent, for it has no long-term structure, mainly when you use the character’s names consistently.
  • The Verge also added that stories tend to introduce and forget characters and other information.

More on OpenAI’s GPT-2 and Its Expectations for the Future

More on OpenAI's GPT-2 and Its Expectations for the Future - THESIS.PH

GPT-2 was only primarily accessible for scientists and individual journalists in the past, which is how Tiago discovered the algorithm. Given this A.I., its technical background, detection for plagiarisms, and other concerns become more challenging for the project team.

Even though there were stated potential risks and criticisms from individuals and experts, the research organization announced their decision to release the algorithm.

They assured that there was no evidence of misuse of the algorithm’s smaller version released through OpenAI’s blog post. Many criticized this; people stated that the research was limited, and Delip Rao also told The Verge that OpenAI spent less than enough time testing its safety.

To address this, OpenAI mentioned that they would continue to observe how the public uses GPT-2 and will develop their policies in publishing their AI research for a first.

Have you ever thought of developing a language modeling system for your thesis?

LEAVE A REPLY

Please enter your comment!
Please enter your name here