Looks Like the Intelligence Explosion Will Happen in 2027 After All (blog.aifutures.org)

A recent METR report suggests that AI research and development at Anthropic has accelerated at an unprecedented rate, approximately doubling compared to historical trends. The analysis estimates that about 80% of this acceleration is attributable to AI, indicating a 1.7x increase in progress driven by AI technologies. This rapid pace could lead to the emergence of Superhuman AI Researchers (SAR) by late 2027, a milestone where AI systems may outperform the world's best human researchers. Additionally, there is a 35-65% likelihood that AIs capable of automating all white-collar work, referred to as Top-Expert-Dominating AIs (TED-AI), could arrive before 2028. The implications of this accelerated progress are significant for the AI/ML community. If SAR arrives as predicted, there is an urgent call for an industry-wide pause in frontline AI development to ensure alignment and safety, given the risks associated with a misaligned SAR potentially leading to recursive self-improvement. Furthermore, the urgency is heightened by concerns that geopolitical factors may allow other countries, particularly China, to replicate or steal advanced AI capabilities. As the pace of AI advancement becomes increasingly unpredictable, these developments necessitate careful consideration of regulatory frameworks and ethical standards in AI research.