Did anyone listen to Geoffrey Hinton's warnings about AI? Apparently not

A note to the reader: the content downloaded from this link is a Medium article marked as a "member-only story" (paywalled) and it cuts off abruptly mid-sentence. We only have the introductory fragment, so this summary sticks strictly to what appears in that visible stretch, without completing or assuming the argument that follows.

The text opens with a recent piece of news: Jacob Coxon, who worked on model training for OpenAI and Anthropic over the past three years, publicly announced his resignation on X on September 9. In his message he warned that both companies are racing "at full speed" toward a superintelligence capable of self-improvement, and stated that this concern is shared by the most senior researchers as well.

According to the fragment, two high-level Anthropic engineers, Evan Hubinger and Samuel Marks, publicly backed Coxon's position, putting at 10% the probability that AI will wipe out humanity within the next decade.

The article connects this to Geoffrey Hinton, who left Google in 2023 after warning about the risks AI poses to humanity. The author recalls that at the time some of those warnings were ridiculed and that he himself, while accepting that risks existed, argued that halting AI development would be unfeasible given the combination of powerful interests and human nature itself. It also notes that in December 2024 Hinton revised his estimate, putting at between 10% and 20% the probability that AI ends up taking control over humanity within 30 years.

The available fragment closes by noting that a growing number of researchers and people working directly on improving the latest AI models are showing deep concern about their capabilities, about the possibility that they surpass human intelligence, and about their ending up making decisions autonomously. The last sentence that can be read — "we have already seen what AI can do when the most basic precautions are discarded, as well as…" — is cut off, so we cannot know what additional examples or arguments the rest of the article developed.