ChatGPT creators struggle to find the antidote

ChatGPT creators struggle to find the antidote

The SME that developed ChatGPT, the successful software that can generate articles as well as poems or essays, is struggling to develop a program that can detect whether a text was written by an artificial intelligence (AI).

• Also read: Snapchat continues to seduce users but disappoints the market

• Also read: Could artificial intelligence steal your job?

OpenAI on Tuesday launched a new open access tool designed to help people distinguish between texts created with AI-based software (like ChatGPT, but not only) and those written by humans.

For example, it must be recognizable whether dissertations were written by a computer program and not by a student, or whether we are addressing a human or a “chatbot” (conversational robot) in an online news service.

But it is currently “impossible to reliably recognize all texts written with AI,” OpenAI warns in its press release from the outset.

According to its own estimates, the new software correctly identifies 26% of all texts written by algorithms as “probably written with AI”.

And 9% of the time, it classifies human-written text as written with AI.

It is “much more reliable” than the previous version, OpenAI emphasizes, but remains “very unreliable” for short texts (less than 1000 characters) and it is better to use it in English.

Irony of the situation aside, the stakes are high.

Many experts fear that generative AI technologies like ChatGPT’s will be used specifically to automate the large-scale creation of scams or highly credible disinformation campaigns.

Similar tools have already been developed with varying levels of reliability, such as: B. GPTZero developed by Edward Tian, ​​​​a student at Princeton University.

OpenAI, a startup co-founded in San Francisco in 2015 by Elon Musk — the Tesla boss left the company in 2018 — launched ChatGPT last November.

The easy-to-use platform presents itself as a chatbot and produces enigmatic texts on simple request. Since then it has been a resounding success, carried over through hypermedia coverage.

The company was previously only known in limited circles for two automated chatting software, DALL-E for image generation and GPT-3 for text generation (ChatGPT is based on GPT-3).

She received $1 billion from Microsoft in 2019 and has just inked a new deal with the billionaire IT giant.