I spent the entire time reading it pondering the same thing.
1. The article presents that calling out to a tool like python is "expensive" because of the overhead of forking a process, loading up the python env etc, but why not just eliminate that overhead and embed WebAssembly so this "tool call" is near zero? This feels very similar to the discussion in the 90's around the overhead of threads v.s. processes or kernel space v.s. user space. Could even go further and have a running beam vm so the LLM can write elixir which is ideal for LLM's that stream out code? Elixir programs will be a lot shorter than webassembly.
2. The core argument stated is "A system that cannot compute cannot truly internalize what computation is." The idea being that it could write a program, execute it and by seeing all of the steps maybe even part way through stop and change its mind or when writing new programs write them better, aka be able to debug on the fly?
3. Not mentioned, but there is a 3rd x factor that LLM's will use this new found computation engine to do overall better at "thinking". Computing in very unexpected ways and to unexpected problems. Maybe it would do dramatically better at some benchmark because of this?
Unfortunately these are not explored and it is just an execution engine even resulting in the conclusion stating "arbitrary programs can be compiled directly into the transformer weights, bypassing the need to represent them as token sequences at all." which goes to point number 1 of if we are compiling to weights why not just optimize the tool calling?
Having worked on software that runs manufacturing plants your comment echos the idea that too many engineers have that they are "better" than manufacturing and lessons don't apply to them.
Go back to your desk and work on a PR that is going to go through a 20 step process that is constantly changing before a hopefully semi-regular release goes out to customers and tell me how you ignoring all of knowledge on how to do this well is good for your career.
For a long time I assumed folks like you were simply uneducated, but know I see it for what it is, elitism.
Depends on the microplastics, but many act as endocrine disruptors that "mimic" estrogen, tricking the body into over-activating ERa and upregulates the PEMT gene and the higher acetylcholine. It could also be that the microplastics can physically bind to or chemically inhibit acetylcholinesterase and that is the reason for the higher acetylcholine. Depending on the cause this is only a short term good thing, but could be downregulating genes.
The non-DHA omega-3 EPA are good at preventing perivascular fibrosis and thus a better glymphatic system for the removal of beta-amyloid proteins. EPA also helps produce melatonin which kick off sleep and this whole process.
Natto-serrazime is probably an excellent complement as it is on the other side and is a dissolver. (Noteworthy: Pterostilbene + Glucosamine similar to EPA reduces fibrosis)
The interesting connection is how this is needed when we are older, but not younger. When younger ERa activates more which does this all on its own. This is the connection to why 2/3 of alzheimer's are post-menopausal women and why HRT is important.
Edit: and to tie this to APOE as it is the gene most associated with Alzheimer's. e4/e4 requires more choline so someone with e4/e4 is more likely to be choline deficient. EPA/DHA usually attach to Phosphatidylcholine (PC) when in the blood/brain. PEMT is a gene controlled by ERa to make choline, but from the above less ERa activation and we make less PEMT so less choline and less PC. Choline is the precursor to Acetylcholine (primary neurotransmitter for memory and focus and essential for REM sleep). This is why Choline is known to help with Alzheimer's.
It depends on the person and their genetics. The further you get from the ERa the more complicated this gets and simply stating "Do X" wouldn't apply to everyone even if there are some incredibly common things to do. I might know all the upstream genes, their interactions, and symptoms by heart so it is pretty easy to identify, but general advice would go something like: eat well, get sleep and exercise.
ERa activation promotes PGE2 resulting in decreased 15-PGDH.
So this is one of those standard poor estrogen signaling downstream things and simply improving the estrogen signaling and you get improved cartilage. Anyone can do this today along with getting all of the other positive effects. Those with EDS who have say variants on their TNXA/B have poor production ability to start and so we do everything we can to improve their cartilage production as they can only make so much which include doing stuff like this.
I am not saying that if there was a pill on the market today that would increase plaque-clearing there wouldn't still be dementia, simply that they would probably then die from pneumonia.
One needs to also dive into neurogenesis. Elevated HPA Axis activation which brought us the oxy-androgens also gives use higher cortisol and lower α-MSH which promote neurogenesis. This elevated cortisol is one of the classic hallmarks of Alzheimer's. Estrogen signaling is also an important part of neurodevelopment. So someone with Alzheimer's usually has low α-MSH and low estrogen signaling and overall reduced neurodevelopment. Need to fix these to prevent dementia.
The various stuff that has been found to reduces dementia also improve estrogen signaling. Estrogens are an incredibly old hormone used in regulating a ton of basic cell function. Keep ERa functioning well and you get not only better amyloid plaque clearance, but healthier DNA repair, immune system, etc and better neurogenesis. A lot of stuff that improves longevity ultimately just keeps this going.
There is a Alzheimer's Choline camp which is where we know that Choline supplements is associated with improved Alzheimer's. With low estrogen signaling you have lower N-methyltransferase so less Choline in the body as it is how the body makes it. Less Choline, less SAM via BHMT, less SAM, less clearance of 2-OHE1, less ERa activation, and ... less choline in a loop. Supplement with Choline and it helps not only with the APOE e4 which consumes more Choline, but undoes this is what this theory is about.
As we are having fun with this, decrease the BHMT from less Choline and it shifts from BHMT-mediated methionine synthesis to MTR-mediated pathways, consuming 5-MTHF resulting in slower THF regeneration. Slower THF regeneration impairs the folate cycle, increasing reliance on serine as a one-carbon donor. Elevated serine flux upregulates PHGDH, which promotes IKKα-HMGB1 signaling which drives amyloid pathology and neuron death! Which brings us to probably my favorite paper from this year. "Transcriptional regulation by PHGDH drives amyloid pathology in Alzheimer’s disease" https://www.cell.com/cell/fulltext/S0092-8674(25)00397-6
I myself don't put myself in the Choline camp, but more see Alzheimer's simply as one failure mode on the metabolic estrogen axis of which is induce in certain genetic or diet configurations. In this mode it results in a number of cascading failures until death. Some genetics/diet are fairly easy to halt/undo the spiral of death which is seen in how there are some things that are known to improve Alzheimer's in some, but not all patients. I can induce Alzheimer's via the gut, diet, inflammation, adrenals, sleep, a whole host of ways, anything that starts the cascade.
An example of an unlucky genetic case is when someone has 3 CYP21A2 rather than 2, they end up with hypercortisolism and nearly always they will get Alzheimer's if there is no intervention. This is a rare genetic case, but simply a way to break the axis elsewhere with the same effect.
For this particular situation on D3 I personally (who is not your doctor) would go with vitamin D3-loaded nanoemulsion. The reason is that Vit D influences how tryptophan is converted down the 5-HTP and serotonin path or the Kynurenine path. We want higher serotonin AND specifically in the brain. The higher serotonin means better melatonin which not only increase sleep, but increase the ERα expression which we are trying to increase... in the brain.
In general: Omega-3, bcomplex with choline etc all have studies. Really it depends on the individual and what their genetic weakest issue is. Its old and boring, but eat healthy, don't eat before bed, exercise (dance!), and get good sleep always apply.
The title is a bit link-bait. It should really be "Disrupting circadian rhythms of plaque-clearing brain cells is associated with Alzheimer's"
> He found that too much of YKL-40, which is linked to Alzheimer’s risk in humans, leads to amyloid build-up, an accumulation that is a hallmark of the neurodegenerative disease.
There are countless studies that highlight how genetics or lifestyle and other factors that result in a reduction of estrogen signaling are associated with Alzheimer's. Estrogen, primarily activated at night decreases the expression of the YKL-40 gene. All of the known interventions, from vitamin D, Mg, to gut, choline, etc all can improve estrogen signaling, decreasing YKL-40 gene. One can end up with Alzheimer's from many different routes so interventions depend on the person.
If there was a pill on the market today that would only increase the plaque-clearing all this really does is move the needle, they still have reduced estrogen signaling and the next weakest part of the system would fail such as from animpaired immune system and they will probably die of pneumonia.
But we could back up and say what is the most common cause of the global reduced estrogen signaling? Often increased oxy-androgens (which increase as we age), so for example 11-ketotestosterone (11-KT) which can't convert to its estrogen form results in upregulates HSD17B2. Why do we have so much inflammation causing increased oxy-androgens from the adrenals? Senescence cells releasing inflammatory factors SASP. More time more time spent on repair resulting in identity loss and mesenchymal drift. All a fancy way of saying we get older and will probably die from whatever weakest part of the system we have genetically. Fix one thing and something else breaks instead.
And for those that want to bring in the most well known genetic mutation APOE e4: APOE e4/e4 has elevated choline demands hindering estrogen signaling as well as raising HDL and lowering LDL. Low estrogen influences Cholesteryl Ester Transfer Protein, raising HDL and lowers LDL beyond what e4/e4 does by itself. With less choline and less phosphatidylcholine, it decreases GLUT1 transporters reducing glucose entering the brain. All of the above leads to an escalating amyloid plaque burden. Then reduced deep sleep and the glymphatic system cleaning is reduced too and you have Alzheimer's.
The above was just from memory probably had an error, but the point is Alzheimer's is not "simple" like this article pretends.
1. The article presents that calling out to a tool like python is "expensive" because of the overhead of forking a process, loading up the python env etc, but why not just eliminate that overhead and embed WebAssembly so this "tool call" is near zero? This feels very similar to the discussion in the 90's around the overhead of threads v.s. processes or kernel space v.s. user space. Could even go further and have a running beam vm so the LLM can write elixir which is ideal for LLM's that stream out code? Elixir programs will be a lot shorter than webassembly.
2. The core argument stated is "A system that cannot compute cannot truly internalize what computation is." The idea being that it could write a program, execute it and by seeing all of the steps maybe even part way through stop and change its mind or when writing new programs write them better, aka be able to debug on the fly?
3. Not mentioned, but there is a 3rd x factor that LLM's will use this new found computation engine to do overall better at "thinking". Computing in very unexpected ways and to unexpected problems. Maybe it would do dramatically better at some benchmark because of this?
Unfortunately these are not explored and it is just an execution engine even resulting in the conclusion stating "arbitrary programs can be compiled directly into the transformer weights, bypassing the need to represent them as token sequences at all." which goes to point number 1 of if we are compiling to weights why not just optimize the tool calling?