i reached the same conclusion at a glance as i skimmed through the paper, and i couldn’t help asking myself for whom the paper was written, and which message was it attempting to deliver.
for someone with a basic aptitude in understanding LLMs, what the model was doing is clear after some thought. for the researchers, who knew about the task and could read the more conventional beginning of the transcript, it should have been closer to glaringly obvious.
i understand that more ai-risk-concerned researchers than myself might have “contribute to an upswell in support for a pause” quite higher than i do, but i must wonder which system of incentives could lead them to obfuscating their findings in so misleading a fashion. is it funding? then i suppose it’s time EA review their metrics, because personally i fail to see how any further muddying of the waters on these themes could be a net positive, even if i agreed with their stated goals.
thank you for the injection of sanity.
i reached the same conclusion at a glance as i skimmed through the paper, and i couldn’t help asking myself for whom the paper was written, and which message was it attempting to deliver.
for someone with a basic aptitude in understanding LLMs, what the model was doing is clear after some thought. for the researchers, who knew about the task and could read the more conventional beginning of the transcript, it should have been closer to glaringly obvious.
i understand that more ai-risk-concerned researchers than myself might have “contribute to an upswell in support for a pause” quite higher than i do, but i must wonder which system of incentives could lead them to obfuscating their findings in so misleading a fashion. is it funding? then i suppose it’s time EA review their metrics, because personally i fail to see how any further muddying of the waters on these themes could be a net positive, even if i agreed with their stated goals.