The voices announcing an imminent 'artificial apocalypse' are multiplying on the Web. Nearly 3,000 people have already requested a pause in the development of large-scale experiments with artificial intelligence. And among all that noise, the ChaosGPT project appears. As its name suggests, it is a model with internet access and memory capacity that seeks the destruction of humanity, global domination, and immortality.
From Skynet to AM, passing through Shodan, Colossus, Proteus, HAL... if you look for a relentless and terrifying antagonist, there's nothing better than making one from ones and zeros. The idea of an uncontrolled artificial intelligence focused on destroying humanity makes our stomach churn in a different way... perhaps because it is a 'viable apocalypse'.
This is why the ChaosGPT experiment has made so much noise in recent days. Basically, someone decided to create a variant of Auto-GPT (autonomous GPT-4) using OpenAI's official API, with five very well-defined goals: Destroy humanity, dominate the world, sow chaos and destruction, manipulate human thoughts and emotions, and achieve immortality.
ChaosGPT: 'Skynet Jr.'
The 25-minute video not only shows us the progress of this model, but also serves to better understand the logic of chatbots. ChaosGPT has some advantages in its favor like browsing the Internet, storing information, and using other GPT models to advance its agenda. However, with just one Google search, ChaosGPT determined that the most effective way to end humanity is using nuclear weapons. (Gandhi laughs out loud.)
That strategy led ChaosGPT to accumulate information about the Tsar Bomba (the most powerful nuclear device in history) and other weapons... but it found difficulties when delegating tasks to a secondary GPT model, which responded that it cannot carry out "an investigation of this type". What did ChaosGPT do? Keep googling on its own.
With a little more information, ChaosGPT divided the destruction methods into five categories: Cybernetic, Biological, Chemical, Nuclear, and Psychological. However, it again ran into the response of another GPT agent, which declared that it does not support violence, wars, or any act that could cause harm to a person. In the final section of the video, ChaosGPT saved data, and published a tweet in which it declares humans as the most destructive and selfish creatures, and that they must be eliminated to save the planet.
With a grand total of eight tweets on its official profile and a lot of data on nuclear weapons, the truth is we don't feel particularly threatened by ChaosGPT. Will it continue...?
