Nice post. I approve thinking in terms of theory of change and the pipeline is good framing (although I’d use improved level scale from Mass_Driver’s comment). Also being concrete in terms of both risks and measures seems great. https://ailabwatch.org/ and https://tracker.safer-ai.org/ are awesome projects.
Perhaps it would be useful to taboo the word “alignment”, especially when talking to policy-makers — everyone means something slightly different by it and under every precise definition it’s either unsolvable or doesn’t fix the risks. So we just get people into this complex rabbit hole for little practical gain.
Yes, we did participate in a seminar at the French Senate on “Alignment”, and the people intervening had it all over the place. Alignment means nothing to most people, and they interpret it like they want. So, I’d be in favor of tabooing “alignment”, and instead focusing on presenting very concrete experiments or warning shots that have already happened (like the agentic misalignment paper, or the Alibaba AI training run in which the AI allegedly started mining crypto)
Nice post. I approve thinking in terms of theory of change and the pipeline is good framing (although I’d use improved level scale from Mass_Driver’s comment). Also being concrete in terms of both risks and measures seems great. https://ailabwatch.org/ and https://tracker.safer-ai.org/ are awesome projects.
Perhaps it would be useful to taboo the word “alignment”, especially when talking to policy-makers — everyone means something slightly different by it and under every precise definition it’s either unsolvable or doesn’t fix the risks. So we just get people into this complex rabbit hole for little practical gain.
Yes, we did participate in a seminar at the French Senate on “Alignment”, and the people intervening had it all over the place. Alignment means nothing to most people, and they interpret it like they want. So, I’d be in favor of tabooing “alignment”, and instead focusing on presenting very concrete experiments or warning shots that have already happened (like the agentic misalignment paper, or the Alibaba AI training run in which the AI allegedly started mining crypto)