Anthropic warns of AI era

Anthropic has warned that AI models are nearing the point of helping develop more capable generations of models, while stressing that its own models have not yet reached full recursive self-improvement. The company said that Claude wrote more than 80% of the code it merged in May, and that its engineers now merge about eight times as much code each day as they did in 2024. In a separate measure, the speedup in an internal software experiment rose from roughly 3× to 52× between two tests. These advances could accelerate research and innovation, but they also increase the risks of errors and biases accumulating and of human oversight struggling to keep pace. That is prompting the company to emphasize supervision, safety testing, and the ability to halt processes if safeguards prove inadequate.

Size: 120%Text Size

Notice: This Is an AI-Generated Summary

Display The Full Article