A startup claims it broke through a bottleneck that’s holding back LLMs

- Subquadratic claims to have solved the LLM quadratic bottleneck, promising SubQ models that are faster, cheaper, and more energy-efficient than the current transformer-based giants. They’ve brought receipts from Appen, suggesting sparse attention isn’t just vaporware. While I usually treat AI breakthroughs like unsealed cargo—full of holes and potential leaks—the third-party validation is hard to ignore. If they can actually process 12x more text without melting the grid, it’s a massive efficiency play. But let’s be real: until it’s widely available and not just a shiny new hash manifest for VCs, keep your expectations in check. This isn’t a moonshot; it’s a potential infrastructure upgrade. Don’t bet the farm on it yet, but don’t sleep on the efficiency gains either.