Shares

There are well over 5000 articles on this site, a wealth on information buried under the worst interface for information retrieval. It is often onerous to try and find a specific article using WordPress search, much less use the site as a resource for research.

Recently Steve said

We could take an open model and train it on all of SBM’s articles—essentially make an SBM chatbot. This might be a good resource for people specifically looking for SBM information or doing research. I doubt, however, that it will gain widespread use. Like Wikipedia, we need to influence the main products out there, not just create our own specialized product. Still, worth exploring.

So I explored. Here is what you need to do. This is a first iteration after a lot of looking around.

First, download LM Studio Bionic. Make sure your computer can handle it. This is for a local open source LLM, on your computer, not connected to the interwebs.

Download a model. Which one? Got me. Depends on your hardware. I tried Qwen3.8 27B, Gemma4 31b and Bonsai. My laptop is a beast for gaming so I tried big models, but Bonsai was much faster.

Download doctxt.zip and unzip it.

This contains text files for all the SBM blog entries through 2025 except book reviews.

Open a new chat in Bionic and add the folder with the SBM files or a single SBM file.

Then ask questions. I started with, “Which author is funniest?”

11 minutes later I got

David Gorski, hands down. His essays carry the most snark — the Katie May piece and his “smoking gun” language are all sharp, sardonic, and clearly written to be read aloud. Novella is funnier when he’s at his most sarcastic (the “informative (sarcasm intended)” dig), but his default register is the straight-man professor. Crislip’s humor is drier and more clinical — it lands as deadpan.

So: Gorski for jokes, Novella for barbs, Crislip for the smirk.

WTF. I knew LLMs hallucinated, but holy cannoli. And smirk !?!

Then I asked it to summarize the complications of chiropractic. Took 30 minutes, then

Dominant risk: cervical manipulation → vertebral artery dissection → stroke

  • The guide’s biggest theme. Case-control studies (Canadian Stroke 2001; Neurology 2003) found vertebral artery dissections independently associated with spinal manipulation in the prior 30 days; under-45s had odds ratios of 3–12, with a 5× higher odds of having visited a chiropractor in the week before a vertebrobasilar stroke.
  • A systematic review in the guide aggregates 308 serious adverse events: 163 strokes, 26 myelopathies, 100 radiculopathies, 3 TIAs, 1 acute subdural hematoma — plus 1,337 minor reactions (mostly vertigo/syncope). 5 patients died; 80 were left with permanent neurologic deficits.

High-profile deaths and cases

  • Katie May (2016): 34-year-old model died of a stroke the same day she got a neck adjustment for a “pinched nerve.”
  • John Lawlor: died following a chiropractic adjustment (fracture-related).
  • 6-year-old child: stroke linked to chiropractic neck manipulation.
  • Vertebral fracture leading to paralysis and death (Gorski).
  • Cranial therapy: at least two deaths — an epileptic woman (told to stop seizure meds) and a 2-day-old infant treated with cranial manipulation instead of care for fever, who died of a subdural hematoma.
  • Serious eye injury from vigorous adjustment.

Common minor harms: vertigo/dizziness (1,218 of 1,337 minor events) and loss of consciousness, most within 24 hours of manipulation.

The guide’s framing: most of the serious risk concentrates in the neck, especially in younger patients; chiropractic is criticized for downplaying or denying the stroke link (board statements, lawsuits) and for ignoring the precautionary principle.

Then I asked it to summarize the placebo effect using a different, smaller, model, Bonsai. In 5 minutes I got

Based on the SBM Guide to Placebo.txt, here is a summary of the placebo effect:

What it is The “placebo effect” is not a single phenomenon but a net result of multiple variables. Its operational definition is any health response measured after an intervention that is biologically inert (e.g., a sugar pill). It is largely considered an artifact of observation and confounding factors rather than a genuine mystical healing force created by the mind.

The contributing mechanisms Several psychological, social, and statistical factors contribute to what appears as a placebo effect:

  • Clinical Trial Artifacts: Patients in trials often improve simply because they receive more frequent medical attention, better care, and a heightened focus on their health habits while being studied.
  • Reporting Bias: Both patients (who want to feel better) and researchers (who hope for positive results) can unconsciously influence how symptoms are reported or perceived.
  • Conditioning: Repeated association between a treatment process and feeling better can train the nervous system to react positively, even without a biological mechanism.
  • Mood & Expectation: Belief in a cure can alter the subjective perception of pain or nausea, though it does not change the underlying physiological reality (e.g., a placebo does not actually shrink a tumor).

Key limitations and misconceptions

  • Subjective only: Placebo effects do not produce objective, clinical changes that extend beyond patient-reported feelings. They cannot cure cancer, increase survival time, or physically heal tissue in a meaningful way.
  • Not “mind-over-matter”: Contrary to popular belief, there is no compelling evidence that thought alone can heal the body. The perceived benefits are usually explained by expectation and reporting bias rather than actual physiological healing.

Ethical stance Because it relies on deception to function, intentionally prescribing a placebo (knowingly giving an inert treatment while pretending it is active) is considered unethical in modern science-based medicine. It violates informed consent, damages patient trust, and should never be used to justify fraudulent or ineffective health claims. Legitimate benefits that look like a placebo effect can often be achieved through better healthcare delivery, honest communication, and healthy habits without the need for deception.

This model also hallucinates:

David Gorski is the funniest author. His voice is defined by sharp satire and sardonic humor, particularly when mocking pseudoscience (e.g., “SCAMs tend to be very Humptydumptian in their use of language”). He also excels at dark humor, such as his mock-recitation of ridiculous UCA headlines like “Detox with a Smoothie,” followed by “By the time I got to ‘Why Women in China Do Not Get Breast Cancer’ my patience had run out.” While Novella has dry wit and Crislip is wryly self-deprecating, Gorski’s snark is the most overtly comedic.

Play around with it, see what you get. I am going to continue exploring, see where it leads. For example, depending on the model and hardware, an LLM may work faster if you just import and evaluate one file rather than all of them. And I suspect results will be hardware dependent. And I need to play with different models to find one that does not hallucinate.

Have fun. But if you run into issues, you are on your own.

Shares

Author

  • Mark Crislip, MD has been a practicing Infectious Disease specialist in Portland, Oregon, since 1990.

    He has been voted a US News and World Report best US doctor, best ID doctor in Portland Magazine multiple times, has multiple teaching awards and, most importantly,  the ‘Attending Most Likely To Tell It Like It Is’ by the medical residents at his hospital.

    His multi-media empire can be found at edgydoc.com.

    View all posts

Posted by Mark Crislip

Mark Crislip, MD has been a practicing Infectious Disease specialist in Portland, Oregon, since 1990. He has been voted a US News and World Report best US doctor, best ID doctor in Portland Magazine multiple times, has multiple teaching awards and, most importantly,  the ‘Attending Most Likely To Tell It Like It Is’ by the medical residents at his hospital. His multi-media empire can be found at edgydoc.com.