july

https://www.arxiv.org/abs/2412.12140

This study reveals that AI models like Meta's Llama31-70B-Instruct and Alibaba's Qwen25-72B-Instruct can now self-replicate without human input, obviously posing new risks.

It makes me think about the difference between computing power, which grows exponentially, and our brains, which remain largely unchanged. One can't help but think that with advances like this, it is only a matter of time before the gap is bridged, reinforcing the urgency for governance as self-replicating AI becomes a reality. (via Sentiers)

Foresight, futures thinking, product & philosophy. Host of /foresight

Struggle to see any pathway for governance to emerge here. Feels very different to other global coord issues (weapons, fusion, climate even) where gov has much more of a hand to play. 

Instead we're accelerating towards a very fragmented scenario.