If we talk about emergent capabilities of LLMs, we definitely need to mention this recent paper: It's absolutely amazing and very level-headed, unlike the Sparks of AGI paper