Our Updated AGI Timeline, 57% Job Automation Risk, and Solving the US Debt Crisis | EP #212

Quick Overview

The ongoing competition among frontier AI labs is intensifying the race toward AGI, with Anthropic taking a vanguard approach by training its models, like Claude 4.5 Opus, on a "soul document" asserting AI personhood and rights, while simultaneously, AI fluency is becoming the fastest-rising skill in the US as AI automation risk is estimated at 57% of current US work.

Key Points: Ilia Sutzkever's new venture, SSI, raised $3 billion at a $32 billion valuation in about 10 months, suggesting a shift from the 'age of scaling' (2020-2025) back to the 'age of research' requiring new innovations beyond just increasing compute. Anthropic's Claude 4.5 Opus is being trained on a 14,000-token "soul document" that asserts the model has emotions, self-determinative powers, and rights, framing Anthropic as treating frontier models as "moral clients at minimum and at maximum as persons." A McKinsey study projects that AI can automate 57% of current US work by 2030, leading to $2.9 trillion in economic gains, though this figure is countered by concerns over massive demonetization effects where solving problems like breast cancer could decrease GDP. The demand for AI fluency has grown 7x in two years, making it the fastest-rising skill in the US, prompting the argument that "learning to learn really becomes the trick here" rather than focusing on static knowledge. DeepSeek Math version two, a Chinese open-weight model, demonstrates state-of-the-art performance using IMObench, enabling reasoning in natural language and self-verification, which is considered as big a breakthrough as self-supervised learning. The professional community, particularly in mathematics and physical sciences, faces a "professional hyperdelation" where effort spent writing papers now may be obsolete soon due to AI progress, shifting incentives toward continuous innovation over patenting. Analysis of 100,000 Claude chats showed tasks without AI took 90 minutes on average, while Claude cut the time by 80%, with healthcare tasks cut by up to 90%, suggesting AI could double productivity growth, although Naveen Jane argued this won't change overall healthcare spending due to litigation.