I remain somewhat surprised reproducing MA isn't a solved problem by all the folks scrounging around for LLM use cases. One would think that LLMs+tools would have a ready-made training set for learning RL policies on tool calls / finetuning models in the history of meta-analyses published so far.

In fairness, I doubt this aspect of my strategy is representative. I'm also engaging in the strategy you'd assumed (although not w/ any particular hope of success, and w/ a careful eye towards poisoning myself). Anyone seen a carefully reasoned piece on the options from a relevant expert/thinktank?

One sad lesson I take from Trump 2.0 is that the majority of Americans are no longer capable of reasoned debate. So I fear any liberal worldview that ties itself to that broken vehicle is doomed. A liberal dictum to entirely avoid such spaces requires very solid justification now more than ever IMO

I understand the argument, but not the basis for believing this has any demonstrable record of success. I suspect we enable the propaganda more by evacuating the space of liberal views than we would by trying to poison the well (e.g., marking “not interested” on every ad, boycotting advertisers etc)

For one thing, it could just be another propaganda platform that we otherwise willfully grant to the right, adding to the seemingly-effective echo chambers they cultivate in various corners of the popular tech/science/lifestyle podcast world, FoxNews before that, and AM radio syndicate years earlier

No… I’m saying that despite not doing so, it may yet have value. To suggest the opposite - that its value derives exclusively from enabling debate amongst reasonable people - is potential folly, maybe one of those high-minded liberal values we need to reconsider in light of liberal losses.

3 reasons: (1) many folks have no idea professors also do research; (2) it may be seen as gauche to ask, and risk forcing a distinction between research faculty and teaching faculty; (3) you can still respond with how you teach (publish) about what your research has found.

I'm open to this but I question which high-minded liberal ideas need rethinking given our dismal performance among the evidently-unreasonable majority of American voters; I suspect that 'the value of a communication platform arises from its ability to engage reasonable parties in debate' may be one.

I confess I haven;t; it's a war of attrition IMO. I also don't think my contributions do much to legitimize the platform so, in the worst case, I see it mostly as a potential waste of time. I'm quote open to being persuaded, just haven't seen a comprehensive/balanced logical argument for vs against.

I also find the math unclear, and hence continue to engage there so as not to prematurely cede the ground. I concede this may not be terribly effective. But, it seems like the math must surely be clearer to you than you suggest, given the (rather ironically) extreme assertion in the OP!

Thanks for this great work! Can you explain a bit more about why you selected the similarity metrics you did? I was not sure whether there was a principled reason to use Spearman for dimensions, cosine for content and Pearson for brain or if this mixed bag of metrics simply worked best.

Very encouraging work; I'm inspired. What happened with the primary outcome (the LPP)? Regardless of whether the study was positive on its primary outcome, it's obvious there's something here worth exploring further. Congrats!