https://en.wikipedia.org/wiki/Tanycyte these are cells that line the 4th ventricle and connect to the hypothalamus. they are sensitive to compounds in the CSF and are a great potential avenue to get CNS effects without crossing the blood-brain barrier.
https://pmc.ncbi.nlm.nih.gov/articles/PMC11340606/ you can use botulin toxin expression to keep tanycytes from doing endocytosis. this blocks leptin transport into the hypothalamus and results in central obesity and insulin resistance...and impaired spatial memory.
i think i disagree on general principles. research is almost always good. claims that a type of research is bad should be restricted to narrow instances where the risk is high, the benefit is small, and the goals can be met through other means -- like gain-of-function research in biology. artificial intelligence is not like that! artificial intelligence is an incredibly broad field -- it's all computational models of cognition -- and of tremendous economic and practical relevance. and of course understanding current commercial AI systems is critical to making them less dangerous, except in the unlikely event that you are willing to, and plausibly can, shut them all down.
also while i'm sympathetic to "don't listen to the details of arguments when you're suspicious of the motives of your interlocutor", that's something you do because of your own limitations, because you don't trust yourself not to be taken in by smooth talk. the strongest reasoners don't need to shut their ears. I totally shut my ears sometimes, because I'm not an ideal perfect reasoner, but in general the degree to which a person needs to do this should count against their credibility.