The legacy of lies in Alzheimer's science

Original Article ↗ Hacker News Discussion ↗

Use of LLMs and Automation to Deal with Fraud

  • One line of discussion proposes using LLM-based embeddings plus supervision from known-fraud labels to remap the Alzheimer’s literature and reweight papers/clusters that look “tainted.”
  • Others respond that LLMs don’t understand truth and that fraud detection is more about images, statistics, and incentives than text patterns.
  • Some suggest a more modest goal: automated flagging of obviously bad practices (mismatched abstract vs. conclusions, irrelevant or retracted citations), noting this is already partly done with heuristic-based tools.
  • A recurring counterpoint: detection already happens; the deeper failure is that institutions, funders, and journals often ignore clear evidence of fraud unless there’s media pressure.

Amyloid Hypothesis vs. Alternatives

  • Several comments argue amyloid plaques may be protective rather than causative, and that overfocus on amyloid (helped by fraudulent work) crowded out alternatives.
  • Others strongly defend the amyloid hypothesis, citing genetic evidence and recent anti-amyloid antibodies (lecanemab, donanemab) that slow clinical decline ~30% in trials.
  • Critics say that “30%” is a small absolute effect on broad scales, with high cost and serious side effects (including brain swelling/bleeds), and may not be noticeable to patients.
  • Proponents reply that (a) trials start late in disease, (b) full arrest would still yield only small numeric changes over 18 months, and (c) larger benefits may appear with earlier, safer agents.
  • Many stress that Alzheimer’s likely comprises multiple subdiseases and interacting processes (metabolic dysfunction, infections, gut microbiota, sleep, vascular issues), so amyloid is at most one piece.

Sleep, Slow-Wave Activity, and Glymphatic Clearance

  • A substantial subthread focuses on sleep—especially slow-wave sleep—as a possible contributor to amyloid clearance and cognitive preservation.
  • Early trials using auditory stimulation to enhance slow waves show improved sleep architecture and some memory/biomarker signals, but small samples and heterogeneous responses make conclusions tentative.
  • Commenters note bidirectional vicious cycles: amyloid build-up may impair glymphatic function and sleep; poor sleep and elevated cortisol then further worsen pathology.

Structures and Incentives in Biomedical Research

  • Multiple comments describe a pattern where flashy but weak or fraudulent findings attract grants, grow large labs, and then control peer review and funding, suppressing competing hypotheses.
  • Proposals include stricter conflict-of-interest limits, caps on lab size/funding, and real independence in reviews; others argue big, well-run labs can be highly productive and that consulting/startups are important for translation and personal solvency.
  • Long, poorly paid training (grad school, postdoc), especially in high-cost areas, is seen as fueling perverse incentives and making fraud or hype more attractive.
  • There is debate over the prevalence of fraud: image sleuths and paper mills suggest a nontrivial rate; some estimate a sizeable fraction of the literature is unreliable, but exact rates are unclear.

Replication, Policing Fraud, and Trust in Science

  • Many call for funding replication and allowing publication of null results to catch fraud and exaggeration earlier; others argue replications are too narrow and don’t fix underlying incentives.
  • Suggested mechanisms include mandatory data/code sharing, automated image checks, and even “bug bounties” for detecting fraud; skeptics point out journals and universities currently have little real incentive to enforce rigor.
  • Analogies are drawn to aviation’s crew resource management: science needs cultural and procedural changes so juniors and outsiders can safely challenge authority and bad work.
  • Underneath is worry that high-profile fraud plus sensationalist coverage risks broad nihilism about science, even though many note science generally self-corrects—eventually, and often too slowly for patients.

Clinical Reality, Diagnosis, and Caregiving

  • Clinically, commenters describe Alzheimer’s diagnosis as a combination of neuropsychological testing and now, increasingly, amyloid/tau biomarkers; there’s disagreement over how “direct” and definitive this is.
  • Several share harrowing personal stories of early-onset dementia, misdiagnosis, financial and caregiving burdens, and how slow institutional support is.
  • Practical advice emerges: early legal planning (e.g., guardianship), pursuing all available support, protecting caregiver energy, and, where available, using specialized end-of-life or dementia support services.
  • Some propose various preventive or therapeutic lifestyle/diet ideas (keto, plant-based diets, multilingualism, coconut oil), but commenters note that, in general, strong randomized trial evidence is sparse or absent, and it’s hard for laypeople to tell hype from robust findings under current conditions.