Thank you for the detailed reply! I'm outside biosecurity, so unfortunately I'm unable to give you feedback on the mobile sampling idea on its merits, but I know what it's like to have an idea that is still only in your own head. I think the hard part isn't the idea, it's finding the right people to discuss it with, develop it together, and tell you whether the direction makes sense. I hope you find them.
Your point about the community is a useful data point. I'll keep it in mind when I decide between the public curriculum and a cohort. I hope to see your updates on how the pilot develops!
Thank you for writing this. It's the only first-person account of the AGI Strategy course I've found.
Your point about the structural blind spot resonated with me. I work as an AI data annotator specializing in Korean, mostly on safety and red-teaming tasks, and I see a similar pattern in evaluations: they assume the criteria work across languages and cultures, but that breaks even in a high-resource language like Korean.
To your open question, I think it becomes strategic whether anyone intends it or not. Frontier labs are concentrated in a few countries, AI is increasingly treated as a national strategic technology, and access can be shaped through export and access controls. Evaluation is where I see this concretely. Most countries will deploy models built elsewhere, tested against criteria written elsewhere, without the means to check whether those criteria hold in their own context. So yes, I'd agree it's strategic.
A couple of small side questions, if you are willing to share: I'm working through the course's public curriculum first. When you applied, did you already have something like the sentinel detection project in mind, or did it take shape during the course? What would someone miss by working through the curriculum alone rather than in a cohort?
Thank you for the detailed reply! I'm outside biosecurity, so unfortunately I'm unable to give you feedback on the mobile sampling idea on its merits, but I know what it's like to have an idea that is still only in your own head. I think the hard part isn't the idea, it's finding the right people to discuss it with, develop it together, and tell you whether the direction makes sense. I hope you find them.
Your point about the community is a useful data point. I'll keep it in mind when I decide between the public curriculum and a cohort. I hope to see your updates on how the pilot develops!
Thank you for writing this. It's the only first-person account of the AGI Strategy course I've found.
Your point about the structural blind spot resonated with me. I work as an AI data annotator specializing in Korean, mostly on safety and red-teaming tasks, and I see a similar pattern in evaluations: they assume the criteria work across languages and cultures, but that breaks even in a high-resource language like Korean.
To your open question, I think it becomes strategic whether anyone intends it or not. Frontier labs are concentrated in a few countries, AI is increasingly treated as a national strategic technology, and access can be shaped through export and access controls. Evaluation is where I see this concretely. Most countries will deploy models built elsewhere, tested against criteria written elsewhere, without the means to check whether those criteria hold in their own context. So yes, I'd agree it's strategic.
A couple of small side questions, if you are willing to share: I'm working through the course's public curriculum first. When you applied, did you already have something like the sentinel detection project in mind, or did it take shape during the course? What would someone miss by working through the curriculum alone rather than in a cohort?