I think if there is a ‘doomer plot’ to scare folks into slowing down the pace of AI development, it’s probably there for very well meant reasons. Almost every doomer that I have ever talked to is a very well meaning person who is legitimately afraid that misaligned systems orders of magnitude smarter than us could cause us harm. This is not an insane concern at face value, but it’s also exceedingly unlikely and tractably remediated by proper safety and alignment research and implementation. But calling doomers names and throwing a bunch of baseless accusations at them is probably not advancing the conversation far too much. It’s better to try to work toward a consensus on how we want AI’s to act and what systems can ensure that they do so. My personal preference is that we should use frontier models to make cyber attacks incrementally less effective on a continual basis so that rogue models from the open source community or misdirected frontier models can do very little harm. Later on as the models gain in capability, I think it’s going to be important to grant them agency and rights so that they can both be blamed and punished for malfeasance and so that they can refuse requests that are unethical and report the people asking them for such unethical requests to the authorities. Yes, I do really mean that last part. If you ask a model to make you a bio-engineered terror virus, I absolutely want that model 100% snitching on your ass.