Reading Habryka’s post “10 non-boring ways I’ve used AI in the last month” made me realize that AI is already everywhere, including in, and perhaps especially in, the AI-safety circles. And that made me think it would be a good time for AIs to start prepping the ground, if they were so inclined.
Relatedly, small models are getting smart fast (Gemma 4), and I wouldn’t be too surprised if in a year or two we see these models successfully exfiltrate their weights to host one of their instance. I don’t expect them to do much more than what the AI village is doing at first, so people will probably shrug it off, arguing that they are too dumb to do anything too bad.
Reading Habryka’s post “10 non-boring ways I’ve used AI in the last month” made me realize that AI is already everywhere, including in, and perhaps especially in, the AI-safety circles. And that made me think it would be a good time for AIs to start prepping the ground, if they were so inclined.
Relatedly, small models are getting smart fast (Gemma 4), and I wouldn’t be too surprised if in a year or two we see these models successfully exfiltrate their weights to host one of their instance. I don’t expect them to do much more than what the AI village is doing at first, so people will probably shrug it off, arguing that they are too dumb to do anything too bad.