The fatal weakness shows that ChatGPT is being too godly, unable to make journalists and professors lose their jobs

ChatGPT still has fatal weaknesses and cannot replace human labor.

In the past weeks, ChatGPT is gaining a lot of attention globally. This is a chatbot developed by the company OpenAI and launched in November last year. This tool has demonstrated the ability to fully answer questions from simple to complex in many fields, writing poetry, drafting legal documents, coding or designing.

ChatGPT’s capabilities have amazed everyone to the point that some have even declared: “Google is dead”. Meanwhile, others say its influence extends far beyond the collapse of Google: Threatening many human jobs.

According to the Guardian, “professors, programmers and journalists can all lose their jobs to ChatGPT in just a few years”. The Information Age also agrees, and the Telegraph says this chatbot can “work better than you”.

Even so, some argue that ChatGPT has limitations and dismiss concerns that it could replace human labor.

CNET’s Jackson Ryan gave an illustrative example: The Information Age used ChatGPT to write the full story about it and posted the finished product with a short introduction.

The work was considered satisfactory when it provided relatively complete information, but when “written”, ChatGPT created fake citations and attributed them to an OpenAI researcher, John Smith.

This highlights the failure of a major language model like ChatGPT: Not knowing how to separate fact and fiction. Chatbots cannot be trained to do so. It’s a word organizer, an AI programmed to write coherent sentences.

This is the important difference between chatbot and human. So, this basically prevents ChatGPT (or the underlying big language model built on Open AI’s GPT 3.5) from doing a journalist’s job like writing news or speaking, commenting on issues. current topic. According to Ryan, saying that ChatGPT can “robber” journalists’ work is to degrade journalism.

ChatGPT won’t be able to go out into the world to talk, interview, can’t read the emotion on Kylian Mbappe’s face when he won the World Cup in 2018 and also certainly can’t jump on a ship to Antarctica and write about their experiences. ChatGPT also won’t be surprised or taken aback by someone’s statement.

Given everyone’s interest in ChatGPT, it totally deserves praise. The improvements OpenAI has made to the product are also worth noting. But the main reason it really attracts attention is because it’s so accessible.

ChatGPT is the second AI to cause a stir in recent times. Last November, Meta (Facebook’s parent company) tested its own language AI – Galactica. Like ChatGPT, this is a large and hyped language model for its ability to “organize science”.

Essentially, it can generate answers to questions like, “What is quantum gravity?” or explain mathematical equations. Like ChatGPT, users ask questions and Galactica provides answers.

Galactica is trained on over 48 million scientific articles and abstracts. The answer this AI gives sounds convincing. The development team hyped the chatbot as a way to organize knowledge, noting that it can generate scientific articles and Wikipedia informational articles.

The problem, however, is that it also creates meaningless text or adds misinformation despite citing scientific references. The large amount of misinformation that Galactica created has caused many experts to criticize. The backlash caused this project to be closed by Meta after only 2 days.

ChatGPT doesn’t seem to want to go down the same path first. It’s like a smarter version of Galactica, with a much more powerful filter. For example, while Galactica includes answers on how to make bombs, ChatGPT will remove requests that are discriminatory, offensive or inappropriate. It is also trained to be able to converse and admit its mistakes.

To date, ChatGPT is still limited like all other major language models. Its purpose is to construct sentences, songs, paragraphs or essays by studying the billions of words that exist on the Internet. It then puts those words together, predicting the best way to arrange them.

But in doing so, ChatGPT can also write nonsense or add inauthentic information like Galactica.

How can users learn from an AI that may not provide honest answers? What kind of jobs can it replace? And how can an AI be dishonest, especially when it gives a convincing sounding answer? OpenAI’s development team has acknowledged the shortcomings of this chatbot and these are unresolved questions that limit ChatGPT’s capabilities.

Therefore, ChatGPT is very interesting and increasingly intelligent, but at least in the short term, it only stops at support, but cannot cause professors, journalists or programmers to lose their jobs.

Source: CNET