Is AI Already Here? AGI's Quiet Arrival

Summary

More than eighty percent of the code shipped by the AI company Anthropic is now written by their own AI, Claude, according to a recent report. This suggests that artificial general intelligence, or AGI, might already be here, not as a far-off science fiction concept, but as a practical reality. The key distinction is between narrow AI, which excels at one specific task, and AGI, which can tackle open-ended problems it's never encountered before, researching and experimenting to find solutions. Anthropic's data shows Claude's success rate on these open-ended coding tasks has jumped from twenty-six percent to seventy-six percent in just six months. Furthermore, the duration of tasks AI can handle independently has grown from a few minutes two years ago to twelve hours recently, with models working for up to sixteen hours straight. While AI is becoming incredibly capable, the report outlines three potential futures: a plateau in AI progress, continued compounding gains where humans guide research, or AI building its own successors, leading to progress limited only by computing power. Anthropic acknowledges that the greatest uncertainty lies in AI alignment – ensuring these systems benefit humanity. They believe a slowdown in AI development would be beneficial for addressing this, but such a pause would require verifiable, global cooperation, which is extremely difficult to achieve. Ultimately, the value is shifting from the 'doing' to the 'directing,' emphasizing human judgment, expertise, and the ability to identify truly worthwhile problems for AI to solve. The gap between individuals and companies effectively using AI is widening, and looking away from its current capabilities is a mistake.