Wednesday, September 30, 21:20
Home security OpenAI's new GPT-2 can be dangerous in the wrong hands

OpenAI's new GPT-2 can be dangerous in the wrong hands


The OpenAI Foundation, founded by Elon Musk 2015, recently released the latest model of GPT-2 text generator, which as it states, can be dangerous in the wrong hands.

Of course, the new model is a little more convincing than its predecessor, so that it can easily fool a man.

The foundation decided to release the first part of the model in February as part of a gradual process. The reason the delayed release of the integrated model was due to concerns about safety, as its creators believe it can be used for malicious purposes, by hacker ή terrorists.

As OpenAI admits, GPT-2 could be used to create misleading articles, online spoofing, automate the production of violent or false social media content, and automate the creation of junk mail and phishing.

Until recently, this model was at the forefront of text creation until BERT was created by Google, which can predict users' searches to bring them more relevant Results.

According to OpenAI, people find text production from the new GPT-2 model of the 1,5 billion parameters "convincing", but slightly more than the 774 million model released in August.

However, OpenAI says GPT-2 can be very persuasively used for malicious use. The CTEC estimates that GPT-2 can be modified to create propaganda that will support white race supremacy, Marxism, Islamism or anarchy.

In one letter accompanying the release of the new GPT-2 model, OpenAI says it believes there are malicious agents out there who have the resources and incentives to use it to their advantage.

However, the institute believes that the risk of lower-level threats, such as economically motivated cybercriminals, is not so immediate. Also, there is no evidence that GPT-2 has been used so far.

OpenAI has developed a crawler model that has about 95% success rates for text crawling. Although this figure is high, it claims that systems detection needs to be enhanced by methods based on metadata, human judgment and public education.


Please enter your comment!
Please enter your name here

Absent Mia
Absent Mia
Being your self, in a world that constantly tries to change you, is your greatest achievement


How to add Sticky Notes to your iPhone home screen

Remember the days when you added sticky notes to your Windows desktop? Well, from now on you can add sticky ...

Vodafone Network Error: Connection Problems Detected!

Vodafone Network Error: Connection Problems Detected! Vodafone has been having problems with internet and telephone connection (internet-telephone) for some time now. According to users ...

As long as the technologies reach their EOL, the hackers are waiting

A recent outbreak of cyber attacks against web commerce sites using Magento 1 underscores the importance of having a strategy ...

Mac: How to export high quality icons from System Preferences

Sometimes, you need a high quality icon from Mac System Preferences for a project, but you can not find any ...

COVID-19 test: Approval in Europe for results in 15 minutes!

A test to detect the Covid-19 virus that gives results in 15 minutes has received the green light for disposal in ...

Gmail: How to add a signature to your emails

Google Gmail supports customizable signatures, which it attaches to every email you send. You can add your name, ...

A spy campaign targets the Indian army!

Security investigators have uncovered evidence of an ongoing espionage campaign targeting India's military (defense and armed forces), ...

Baidu: Malicious pop-up redirects traffic

Malicious pop-ups redirect the traffic of the website of the technology company Baidu, to the intruder's domain. The malicious ...

LinkedIn hacker Nikulin was eventually sentenced to 7 years in prison

A US judge on Tuesday sentenced Yevgeniy Nikulin to seven years in prison, ending an international legal drama ...

Inova Health: another victim of the Blackbaud breach

Inova Health Systems has sent letters to its patients, in order to inform them about a cyber attack, which is possible ...