06-reference

moonshots ep215 gpt52 corporate collapse

Fri Dec 12 2025 19:00:00 GMT-0500 (Eastern Standard Time) ·reference ·source: Moonshots Podcast ·by Peter Diamandis
gpt-52benchmarksknowledge-workgdp-valarc-agihyperdeflationcorporate-collapsepost-training

Moonshots EP 215: GPT 5.2 Release, Corporate Collapse in 2026, and 1.1M Job Loss

Summary

Emergency pod recorded the day after GPT 5.2’s release. Alex walks through benchmarks in detail, identifying three levers OpenAI likely pulled: more compute, safety relaxation, and post-training on specific benchmarks. The standout metric is GDP-val (Gross Domestic Product Eval), where GPT 5.2 scores 70.9% across 44 occupations and 1,320 knowledge work tasks — meaning 71% of comparisons between human and machine resulted in the machine performing better, at 11x the speed and less than 1% of the cost. Alex declares: “knowledge work is cooked.” On Frontier Math Tier 4 (research-grade problems taking mathematicians weeks), GPT 5.2 still trails Gemini 3 Pro (14.6% vs 19%), which Alex frames as a loss for OpenAI despite having sponsored the benchmark’s creation. ARC AGI 2 sees a massive jump from 17.6% to 52.9%, suggesting heavy post-training. The cost of intelligence shows 390x year-over-year hyperdeflation on ARC AGI visual reasoning. Dave notes the practical gap: legacy codebases in Java/C struggle to benefit, while Python rebuilds from scratch are transformative. Colossal Biosciences and de-extinction get a fun opening segment.

Key Segments

Notable Claims

Guests

RDCO Mapping