An artificial intelligence research firm once backed by Elon Musk has announced a tool that can write as well as people, and anyone who tries it compliments it.

OpenAI, one of the major artificial intelligence laboratories in the technology village, has just announced a tool called “natural language generation” called GPT-3. They recently submitted it to a few pre-designated content developers to try it out. According to initial estimates, the GPT-3 paragraphs created almost like artificial products, and everyone was surprised at the ability to write code, imitate famous writers, even GPT-3 even has The second half of a draft of “how to run a board meeting” could be finalized.

GPT-3 is an algorithm that is able to deduce patterns that appear in a huge amount of data of the writer’s text, and then rely on it to draft words and paragraphs similar to the pattern. have found in the database.

Illustration of Alex Castro.

Language creation models have appeared in many places and applied in many systems. For example, Google’s Smart Compose is still helping users perfect the words in the emails they compose, and the Associated Press has been using AI to write sports news. However, the style of the tools is still forced, unnatural due to the limited amount of data for AI training as well as the incomplete language model.

Historically, natural language generation systems still lack many expressions of expression“, Says Carolyn Rose, a professor at Carnegie Mellon University’s Language Technology Institute. But GPT-3 has a different shape. Based on the perceptions of early adopters, GPT-3 surpasses the old natural language generation models thanks to its huge data and 175 billion parameters – the ground rules upon which the algorithm is based. make a decision on which word is most appropriate for the contextual. The previous GPT-2 version had only 1.5 billion parameters, and Microsoft’s technology industry’s strongest natural language model has only 17 billion parameters.

A similar Google search engine called Ask Me Anything – Ask Me Anything, powered by GPT-3.

OpenAI announced the GPT-3 specification in May, but it was not until late July that it would send the tool to designated content developers. On Twitter, GPT-3 testers expressed their surprise: the tool was able to code a website based on English descriptions, writing a single song in Shakespeare style. They describe GPT-3 as a groundbreaking tool that can make job more easy, but also has the potential to make many people lose their jobs.

Positive feedback and negative comments

German artist Mario Klingemann included samples of various authors, and GPT-3 produced meaningful stories with similar writing style.

People consider it a messenger to bring about major changes in natural language processing or coding, depending on their area of ​​expertise.”, Oren Etzioni, CEO of Allen Research Institute. This tool is built on the breakthroughs of its predecessors as well as 30 years of AI research and testing. “It is not clear if this is a big step forward, but it is still confirmed that it is very impressive”, Etzioni said.

Just type a description, the system using GPT-3 will create a site layout (near) exactly what you want.

Because OpenAI publishes GPT-3 as an API, anyone who has limited circumstances – no computational power or expertise in using and training models that generate natural language – is able to use it. use new tools. Although GPT-3 is not a new breakthrough, it can still be a bridge leading to major improvements in the array of automatic word generation. For example, text-to-speech reading systems will be much more pleasing.

Clearly GPT-3 is not perfect, OpenAI CEO Sam Altman must emphasize that it still has many weak points and still make silly mistakes. Although artificial intelligence will change the real world, the GPT-3 is only a small part of that future vision. OpenAI “There are still many aspects to be explored”.

The initial success of the GPT-3 raises old questions about the ethical elements found in an injustice system without knowing what is right and wrong. Mr. Etzioni also noted that GPT-3’s very good ability to fake voice could make the process of filtering fake content harder than before. Not to mention machines will automatically follow and exaggerate the prejudices that exist in society, the prejudices about a small group of people that equate the majority.

After all, this process is all mechanical. It does not determine its value but can not recognize what is beauty”, Professor Carolyn Rose said.

Refer to Business Insider

[ Æsir Tales ]
Back to top button