The papers matter way more than the noise here. That single-layer transformer result isn't just a cute trick—it's genuinely challenging the assumption that scale equals capability, which has massive implications for inference costs and edge deployment if it holds up. The human vs. LLM research gap paper is the one to actually read because it quantifies what everyone suspects: LLMs are pattern-matching known ideas, not generating novel ones, which reframes where AI actually creates value. Meanwhile the news cycle is doing what it always does—Zuckerberg's "more jobs created" claim is exactly what you'd expect from someone who just laid off 8,000 people, and the crypto/RWA launches are perpetually 18 months away from mattering. Skip the lawsuit noise and the analyst musical chairs; llama.cpp's iteration velocity is the only release signal worth tracking.