Daniel Kokotailo, a former researcher at OpenAI and currently the head of the AI Future Project, recently issued a strong warning. He pointed out that the current artificial intelligence industry is caught in a blind competition, with companies racing to build systems they have not yet fully understood or controlled.

Kokotailo emphasized that the core challenge facing the AI industry is the "alignment" issue. Even though AI has surpassed human capabilities in many areas, researchers still cannot fully understand how advanced models make decisions.

The Threat of Superintelligence

This uncertainty makes it extremely difficult to ensure that future AI systems reliably follow human instructions. Kokotailo stated that while the difficulties of implementing AI alignment are an "open secret" within the industry, there is currently no truly viable solution.

He predicts that once humans create superintelligence, humans will no longer be the default dominant species on this planet. Through his analysis of the development speed of AI systems, he found that if companies continue to prioritize speed over safety, unpredictable risks will multiply.

Underestimating the Development Speed

Many people still underestimate the pace of AI evolution, considering discussions about失控 more like science fiction. However, global tech giants are continuously investing billions of dollars in building larger data centers, aiming to develop even more powerful models.