In Lakatosian terms, the amyloid hypothesis is an example of a degenerating research program that has largely failed to predict new observations and is primarily driven by post hoc reasoning. The hypothesis was rescued by research claiming a significant new observation that was ultimately shown to be fraudulent (https://pmc.ncbi.nlm.nih.gov/articles/PMC12397490/).
From a Lakatosian perspective, the amyloid hypothesis is not necessarily wrong, but it is not paying off in terms of empirical insights relative to the amount of attention and funding it has received.
My CEO did a deep dive into AI prototyping and eventually ran into a wall with data architecture and deployment. Fortunately, he realized very quickly that having human designed core infrastructure is what enables vibe coding that doesn't run off the rails.
I'm a non-Conformist is the historic sense. In the past decades, American Christianity has devolved into a smorgasbord of personality cults and group therapy that effectively suppresses the freedoms that many Christians fought for at the country's founding. Not usually overtly, but the prison is in the mind (or its lack of use).
Computational biologist with a focus on predicting individual human health at a startup, but I have ended up managing software engineers. (Scientist explore and engineers make the science work in production.)
The mind-body link is too important to get the causality wrong and The Body Keeps Score is an ideology where the causality only goes one way.
I have a cousin that had frequent, overwhelming anxiety attacks. She started eating breakfast consistently and the anxiety disappeared at the same time. Anxiety is strongly linked to gut activity, so the temporal correlation is a smoking gun, even if not dispositive.
For her, "understanding past trauma" was irrelevant to the solution.
I think you are getting at the need for tiered layers of abstraction and constraint. Simultaneously considering all possible ways to solve a problem doesn't work for humans or the LLMs derived from our use of language. The repeated use of Domain Specific Languages (DSL) in the context of a general purpose programming language gets at this same need to constrain solution spaces within a reasonable boundary.
Once we have quantum LLMs, the need for intermediate abstraction layers might change, but that's very [insert magic here].
Available data makes causality hard to get right. This paper is trying to get around known constraints with observational data (e.g., some people stop drinking when they start having noticeable problems). Mendelian randomization tries to infer how much a person drinks from their genetic variants. However, the genetic tendency to drink might be associated with the same variants related to dementia. The summary doesn't make it clear if this was addressed.
Currently, I'm using generative AI of various kinds on my M1 Air (llm, image gen, TTS, STT), but am frustrated by the limitations - primarily memory and secondarily availability of an MLX adaptation.
Just an AI hobbyist, so I don't have time or inclination to tweak everything. Given the non-NVIDIA GPU, how painful will it be to play around with new AI models on this system?
Natural selection encodes adaptive responses to the environment in DNA (and other molecules), so memories can be encoded to the extent that they are adaptive and can be encoded (i.e., mechanisms may not exist to encode everything using only standing natural variation).