Is AI Progress Real? Four Independent Metrics Show It (skepticcto.substack.com)

A recent analysis by SkepticCTO reveals significant advancements in AI capabilities, substantiated by four independent metrics. Following a flurry of high-profile model releases between November and December 2025—including xAI's Grok 4.1, Google's Gemini 3, Anthropic's Claude Opus 4.5, and OpenAI's GPT-5.2—there's noticeable improvement across multiple benchmarks, debunking skepticism about AI's progress. These models showed unprecedented performance gains around the same timeframe, particularly in task execution, cognitive testing, and logical reasoning, suggesting a genuine acceleration in AI development. The analysis underscores four key mechanisms driving this advancement: pretraining efficiency, enhancing reinforcement learning, developing effective harnesses around models, and leveraging a flywheel effect where improvements feed into one another. Notably, AI models are becoming increasingly capable with less computational resource, thanks to breakthroughs like mixture-of-experts architectures and scalable reinforcement learning techniques. However, challenges remain, particularly around data limitations and reliability in model performance. Overall, the evidence supports that AI's progress is not just real but potentially accelerating, marking a pivotal moment for the AI/ML community.