No point in arguing with people who bring a snowball into Congress to disprove global warming.
922 karma · joined June 24, 2023
Also, I’m not that familiar with the whole process, but I just wanted to say that I think you were too hard on yourself during the last performance drama. So thank you again and remember not to hold yourself to an impossible standard no one else does.
When enough words, framing, and unstated important premises are omitted, it crosses over from the realm of incomplete or misleading into plain outright lying in my worldview.
They claim "Reverses advanced AD in mice." What they did is reverse symptoms in genetic models.
They claim to "Restore NAD+ homeostasis," ignoring that NAD might not even be causally related to Alzheimer’s, just a side effect. It’s like saying we cured a house fire because we efficiently removed the ashes after the fire. It’s the Tau thing all over again.
The claim: "Conservative molecular signatures" when in reality, 5xFAD mice are poor predictors of human clinical success to such a degree that it’s statistically more common for mice studies to NOT transfer to human biology than to do so.
They also make unsupported claims like "Safer than NAD+ precursors (supplements)," when this is a pre-clinical assumption. No human toxicity trials are mentioned in this context, and there are always MASSIVE differences when switching to real human studies. It might be correct, but there’s no basis to say that based on this study.
Also, the senior author owns the company. The paper has the hallmarks of a "pitch deck" for the drug.
In short, it seems to me that the claim of "Full Neurological Recovery" is misleading to patients. It fails to prove that fixing NAD+ in humans will stop the disease, only that it works in mice engineered to have the disease, and only by assuming that their specific measure is a 1-to-1 with the clinical presentation of the disease. The results are likely the "best case scenario" presented to support the commercialization of P7C3-A20.
Here is the COMMON SENSE question peer-reviews should have asked. Is low NAD+ the fire, or just the ashes?
Why should we believe this works in humans when the last 500 'cures' in 5xFAD mice failed?
Are you regrowing a brain, or just cleaning up a dirty one?
How does one molecule fix five unconnected problems simultaneously? The Context: The drug fixed inflammation, blood-brain barrier, amyloid/tau (protein folding), and memory (neuronal signaling). Drugs rarely hit four distinct biological systems with 100% success....
Where is the toxicology report that proves 'safer than supplements'?
Edit: Read through it and have come to the conclusion that the post is 100% OK and properly framed: He explicitly says his approach is to "sharing early and making a fool of myself," prioritizing transparency and rapid iteration over ironclad verification upfront.
One could make an argument that he should have cross-compiler checks, independent audits, or delayed announcements until results are bulletproof across all platforms. But given that he is 100% transparent with his thinking and how he works, it's all good in the hood.
Funny that you ignored “indicates,” which is a perfectly valid inference here, and also it’s true.
If you want to be pedant, don’t self-own.
An energy density of 1270 Wh/L is indeed roughly double what is currently found in top-tier electric vehicles. However, as with many battery research avenues, it is not viable on a practical level unless a major breakthrough is discovered in addition.
Here is a list of all the issues that must be resolved before such battery technology is viable for commercial use.
It only lasts about 100 charge cycles before degrading to 80% capacity, which is not sufficient for commercial use. LiFePO4 reaches this after a minimum of 3000 cycles.
It uses silver. In addition to this likely being a deal-breaker for mass production, the paper probably downplays the mass loading of silver required to maintain that 99.6% efficiency.
Anode-free batteries have zero excess lithium. Every time you charge/discharge, you lose a tiny fraction of lithium to side reactions. The paper claims a Coulombic Efficiency of 99.6%. The fact that they hit ~82% suggests the degradation is severe and inevitable without a massive reservoir of extra lithium, which defeats the "energy density" gain.
Density suppression for 100 cycles is not proof of safety. Dendrites often grow slowly and trigger short circuits later in life (cycle 200+).
There is also the known problem with pouch cells and significant volume change ("breathing"). The paper quotes volumetric density including packaging, but does it account for the swelling that happens after 50 cycles? Often, these cells puff up like balloons, rendering them unusable in a tight battery pack.
They tested at 0.5C (2-hour charge). Fast charging (15-20 mins) typically destroys lithium metal anodes instantly by causing rapid dendrite growth. This technology is likely limited to slow-charging applications.
Finally, there is no mention of temperature effects on performance.
I don’t mean to be negative, and research like this is extremely important. But this research paper is not properly framed. It’s like an archaeologist finding a buried house and extrapolating that this could mean we found an entire city! Why can’t we just say that the archaeologist found an interesting house?
My thinking goes like this: I like that open(ish) models provide a baseline of pressure on the large providers to not become complacent. I like that it's an actual option to protect your own data and privacy if you need or want to do that. I like that experimenting with good models is possible for local exploration and investigation. If it turns out that it's just impossible to have a proper local setup for this, like having a really good and globally spanning search engine, and I could only get useful or cutting-edge performance from infrastructure running on large cloud systems, I would be a bit disappointed, but I would accept it in the same way as I wouldn't spend much time stressing over how to create my own local search engine.
My most important takeaway is that, in theory, I could get a "relatively" cheap Mac Studio and run this locally, and get usable coding assistance without being dependent on any of the large LLM providers. Maybe utilizing Kimik2 in addition. I like that open-weight models are nipping at the feet of the proprietary models.
And a note: EMDR works MUCH better than blue-light-reducing therapy. It's just that the theory for WHY it works is insane (integration of memories/thoughts across brain hemispheres is facilitated by moving eyes back and forth). It's just exposure therapy, and the "follow the light" stuff is just structuring the exposure setting. You get the same effect while doing exposure therapy while driving a car.
Same reason I have been skeptical towards dark energy, EMDR, and the blue light destroys sleep craze. And many other stupid stuff. If you like a story or a finding, that’s a clue to double the critical sceptisism.
This might be fun for vibecode to just let it go crazy and don't stop until an MVP is working, but I'm actually afraid to turn on agent mode with this now.
If it was just over-eager, that would be fine, but it's also not LISTENING to my instructions. Like the previous example, I didn't ask it to install a testing framework, I asked it for options fitting my project. And this happened many times. It feels like it treats user prompts/instructions as: "Suggestions for topics that you can work on."
Skatval is a small local area I live in, so I know when it's bullshitting. Usually, I get a long-winded answer that is PURE Barnum-statement, like "Skatval is a rural area known for its beautiful fields and mountains" and bla bla bla.
Even with minimal thinking (it seems to do none), it gives an extremely good answer. I am really happy about this.
I also noticed it had VERY good scores on tool-use, terminal, and agentic stuff. If that is TRUE, it might be awesome for coding.
I'm tentatively optimistic about this.
My gut intuition just didn't like the framing. Now that I have read through it thoroughly, my answer is this: It's untrustworthy because it is obviously extremely selective with what it includes, omits relevant base rates, uses graphical examples out of context, and has an obvious bias and agenda. That is just one of tens of examples in the article.
Your tobacco reference can be condensed into: "Large firms are known to lie and cover up things." I agree 100%. They plainly outright lied directly AND lied by covering up. But the reaction to that is not to lie better. And by better, I mean lying by omission, juxtaposition, and framing. These are still methods of lying, just that they are harder for people to detect.
I only care about evidence that proves that it causes Parkinson’s, with basic scientific rigor. I’ll eat my hat if any of the cited studies did basic attempt at falsification.
Should we ban anhydrous ammonia, chlorine, or gasoline? They are nasty and dangerous too. The article is purely scare-mongering to make it seem true while obviously pushing an agenda. This is not science. See my reply at the same level after I did a review.
I’m not saying pesticides are health tonics, but this piece feels like pure litigation PR rather than an actual investigation. It prioritizes storytelling over science and engages in what I can only describe as lying by omission. Here are the main issues I found:
The nurse whose skin peeled off just from touching a patient’s urine? The article frames this to make you think, "Wow, this stuff is so toxic that if a farmer uses it, his body becomes a weapon." I looked into the medical case this is likely based on. That patient didn't just "farm" with Paraquat; he ingested a lethal, concentrated dose (usually a suicide attempt). By leaving out that the patient drank a cup of poison, the author conflates Acute Poisoning (death in days, acid urine) with Chronic Exposure (trace amounts over years). If the farmer in the main story had enough Paraquat in his system to burn a nurse’s skin, he wouldn’t be alive to give an interview about Parkinson’s. He’d be dead from multi-organ failure. Omitting this context is manipulative fear-mongering.
Then there is the math: Parkinson’s affects about 1% of the elderly population. There are 2 million farms in the US. Even if Paraquat was essentially harmless water, you would still have tens of thousands of farmers with Parkinson’s purely by chance. The article ignores this base rate to imply that every diagnosis is a result of the chemical. It treats a probabilistic risk as a deterministic cause.
It also ignores confounders (like the "Rural Cluster" Problem). Farming is a "chemical soup" lifestyle. You have well water (a known PD risk), head trauma risks, and exposure to dozens of other chemicals like Rotenone or Maneb. The article presents a direct line: Paraquat -> PD. But scientifically, isolating one chemical from 30 years of rural living is a nightmare. The article doesn't even attempt to falsify the hypothesis or look at other factors; it just assumes the lawsuit's narrative is the scientific truth.
The article also fails basic science standards. It is storytelling, not science. A real scientific inquiry follows Popperian standards—you make a conjecture and then try to disprove it. This article does the opposite: it acts like a defense attorney. It stacks up emotional anecdotes and selective correlations to confirm its bias and ignores the replication crisis in epidemiology where results often don't stick.
This isn't journalism and it’s not science; it’s advocacy via outrage. It uses the real tragedy of these farmers to push a specific narrative, relying on readers not knowing the difference between drinking poison and spraying crops. If you’ve ever wondered why science doesn’t make more progress, and we have the replication crisis, look no further.
“Even secondary exposure can be dangerous. One case published in the Rhode Island Medical Journal described an instance where a 50-year-old man accidentally ingested paraquat, and the nurse treating him was burned by his urine that splashed onto her forearms. Within a day, her skin blistered and sloughed off.
So if we are viewing this through the needle in hey stack lens: The needle was very surprising for the base model, so going forward, when it see anything of the same nature, the memory module will not just give you hay, but the needle, because it made a special note of it when it went through the haystack 1 million tokens ago, because the needle was surprising.
The Transformer's normal attention mechanism is already secretly trying to be a long-term memory system. Every time it writes a new KV pair into the cache, it’s desperately trying to “remember” that token forever.
But it’s doing it in the dumbest possible way: by hoarding an ever-growing pile of raw vectors, then frantically dot-product searching through the pile every single step. It’s like a hoarder who never throws anything away and has to rummage through mountains of junk to find the one receipt they need. Of course it chokes at long contexts.
Titans/MIRAS looks at that mess and says: “Why store memory in a growing garbage pile of vectors? Store it in the weights of a deep neural network instead — and let that network keep training itself in real time, but only on the stuff that actually surprises it.” That’s literally it.
Using the Tim Cook Martian example: The model is cruising through boring financial numbers → attention is doing its normal thing, KV cache is growing, but nothing is really sticking.
Suddenly: “Tim Cook is a Martian.”
Normal attention would just add one more KV pair to the pile and pray it doesn’t get drowned out later.
Titans instead goes: “Holy shit, reconstruction error off the charts → this does NOT fit my current memory at all → massive gradient → actually rewrite huge chunks of the memory MLP’s weights right now so this fact is burned in forever.”
From that moment on, the memory MLP has physically changed its internal wiring. Any future query that even vaguely smells like “Tim Cook” or “Martian” will make the activations explode through the newly rewired paths and spit out a vector screaming “MARTIAN” at the frozen attention layers.
The frozen attention (which is still doing its normal job on the short window) suddenly sees this one extra “virtual token” in its context that is confidently yelling the surprising fact → it attends hard to it → the model answers as if the Martian revelation happened one token ago, even if it was 2 million tokens back.
It looks exactly like a super-attention mechanism that only “primes” or “locks in” the surprising needles and deliberately forgets or ignores the hay. And it is also a way to fine tune one the fly permanently for the current context.
I think…
Basically, my take is: It’s not a technical monoculture; it’s a billing psychology + inertia culture.
I dont think the internet is fragile simply because Cloudflare is so ubiquitous, because that view ignores the economic factor of why people choose them. The situation is really a perfect bi-modal distribution: at the low end, you have hobbyists and personal sites who use Cloudflare because it is the only viable free option, and at the extreme high end, you have massive enterprises that truly need that specific global capacity to scrub terabits of attack traffic.
However, I think the following perspective is important: For the vast middle ground of the internet—most standard businesses and SaaS platforms—Cloudflare could be viewed as redundant. If you are hosting on AWS, Google Cloud, or Azure, you are already sitting behind world-class infrastructure protection that rivals anything Cloudflare offers. The reason this feels like a dangerous monoculture isn't because Google or Amazon can't protect you, but rather because Cloudflare wins on the psychology of billing. They sell a flat-rate insurance policy against attacks, whereas the cloud giants charge for usage, which scares people.
Ultimately, the internet isn't suffering from a lack of technical alternatives to DDoS protection, nor is Cloudflare a NECESSARY single point of failure; it is just suffering from a market preference for predictable invoices over technical redundancy, and inertia, leading to an extremely high usage of Cloudflare. So basically: Even though we are currently relying a lot on Cloudflare, we are far from vendor lock-in, and there is a clear path to live without them, given that there are many alternatives.
Maybe we could view this as a good thing, since basically medium to large-scale enterprises efficiently subsidize small and hobby-level actors? So to summerize: The 2018-era "just use Cloudflare for everything" advice is outdated, and the following is a better philosopy: If you're tiny: Cloudflare free tier is still a no-brainer. If you're huge and actually get attacked: pay for Cloudflare Enterprise or equivalent.
If you're anywhere in between: seriously consider whether you need it at all. The hyperscalers are good enough, and removing Cloudflare can actually improve your availability (fewer moving parts).
I think Cloudflare thinks this way too, which is why they've been pushing Zero Trust, Workers, WARP, Access, and Magic Transit, to become the default network stack for companies, not just the default firewall.
/wall-of-text
Or should I say... Everybody thinks that titles using the format "everybody thinks X" would be more honest if they instead said: "I believe X."