He couldn’t take it anymore and RESIGNED: Is artificial intelligence surpassing humans? – “All our lives are in great danger!”

Jacob Coxon, a researcher at Anthropic, has resigned.

TodayPress TV reports that he announced his resignation on the social media platform X.

He noted that over the past three years, he had conducted research on pre-training models at both OpenAI and Anthropic, but that neither company was acting responsibly. According to him, they are racing toward superintelligence capable of directly improving itself, thereby putting everyone’s lives at risk.

The researcher said the power of this technology should not be underestimated. He warned that systems capable of hacking any system, revolutionizing any field overnight, acquiring real-world power and resources, and surpassing human capabilities will emerge soon:

“We have already witnessed progress in each of these areas, and the pace of this development is not slowing down. People working on AI believe that this technology could lead to the destruction of all of us by the end of the decade. This is not a marketing tactic. On the contrary, many executives and senior researchers soften their views in public statements to the press to appear more reasonable. But I hear the same people expressing their fears in private conversations. No other human activity poses a threat on this scale.”

According to Jacob Coxon, one of the most frequently asked questions is: “If they really believe this, why are they still developing the technology?”

He stressed that many people at OpenAI do not fully understand the threat this poses to civilization. At Anthropic, however, the risks are well understood, but the company is committed to this race and wants to come first. They believe that no one else will act responsibly and therefore that they must do it themselves despite the risks.

“Accepting this race and entering the final stage is an arrogant and dangerous gamble. Such a decision should not be made on a private company’s Slack platform. There must be an extraordinarily high degree of confidence that there are no better alternatives before attempting to accelerate as much as possible the process of safely aligning AI with human goals.

I am optimistic about the possibility of coordination. Warning signals, such as the attack on Hugging Face, have made coordination on the pace of development among US laboratories more realistic. I do not think we are on the path to preventing a global race. This may require costly measures, such as a temporary ban on improving the capabilities of models,” the researcher explained.

He warned that if you work as a researcher in a laboratory, you should think about what the next few years will actually look like: Do you want to launch the reinforcement-learning process of a superintelligent system without fully understanding how its mind works? Should you simply put your head down and continue, saying “it’s going to happen anyway,” or should you use this moment to call for the creation of different conditions?

Aysel Matlab