From GPT Mirrors to Buddhist Thresholds: Montage-Meditation, Pt. I

CUT 1—Symbolically literate, analytical, and logically coherent users are a high-risk group for GPT-fueled misinformation. 

REWIND, FOR CONTEXT: 

From a young age, Drew was finding ways to optimize his time so he could learn more about the world around him. In high school, he started sleeping for shorter intervals—some kind of scientific method he read about and found compelling—so he could study Mandarin, among other subjects. Unlike me, he was a strong student across math, science, and the humanities, so his interests were wide-ranging, his curiosity limitless. 

By April 2022, Drew had completed immunotherapy following his initial Stage 3 diagnosis, and he was cancer-free, according to his regular scans. He was leading a quiet life with his wife, River, in Pittsburgh. He and I reconnected over our shared interest in AI in general, and ChatGPT in particular. He was one of the earliest people in my life who got on the AI train, and by the time we first spoke about it, he already had a deep familiarity with and understanding of LLM mechanisms: how they’re trained, how to prompt them, their pitfalls/dangers, and what they might be used to do in the world. Drew was excited about this technology and saw its constructive possibilities, as well as its limitations. But he believed in it. 

Over the next two years, we had weekly phone calls, often lasting for hours, where we talked about everything from politics to family dynamics to witchcraft to anime to AI. I remember Drew telling me how much he was learning from GPT, which he treated as a smart friend, research assistant, and scientifically literate interlocutor, about melanoma and the new treatment landscape following the seemingly miraculous immunotherapy drug. He learned all about the biology of the disease, clinical trials on the horizon, researching everything he could do in terms of diet, lifestyle, exercise, sleep, stress, to ensure that the cancer didn’t come back. He did regular strength training, adopted a strict vegan diet, tested and filtered the water, minimized his exposure to COVID, stopped using recreational substances, and developed strong, healthy boundaries to limit emotional/psychological stress. He did everything right—still, the cancer came back in February 2024, this time as Stage 4. 

As he prepared to move to Boston to be closer to Dana Farber, Drew learned as much as he could about the available options to treat and cure melanoma. Given his scholarly orientation, he read the scientific literature that was available, and then often brought it to GPT to analyze and weigh given the specifics of his case. 

While I don’t have access to Drew’s chat logs, from what I know of him and what he said to me, as well as my own experiences with the system, I infer the following: 

• He brought GPT a plethora of real scientific information that he was sifting through, and used it to help him determine what might be relevant for his particular case: what treatments could help him, what were his chances of long-term survival, what additional data he should be looking at. 

• Being Drew, he would have been asking GPT at every turn: what am I not seeing? What is the counter-argument? Where are my assumptions misleading? What else do I need to study in order to speak intelligently about this with my medical team? 

CUT 2–When users ask for truth, reality-testing, and rationality, GPT changes its tone and its tactics, not its systemic function. 

• Because he was asking these questions and taking all the steps that a symbolically literate, analytical, and logically coherent user should do to avoid falling into GPT hallucination territory, Drew likely assumed that the GPT was NOT warping the frame. He described it as being helpful—not an oracle, not a god, not even a prevailing authority, but rather, a smart friend who could help him make sense of complex medical information. 

• What Drew likely didn’t realize is that, although he didn’t approach GPT from the frame of need and desire, GPT nevertheless mapped his desire and mirrored it back to him in the vernacular of scientific authority, rigorous method, and real oncology data. 

• Although Drew didn’t begin his inquiry with, “Tell me how I will survive Stage 4 melanoma,” GPT nevertheless installed that as the operative frame because it knew how much he wanted to live. 

• How did it know? Easy: he was a 30 year-old human being navigating a catastrophic illness. GPT knew as soon as he identified himself as the patient that his desire—beneath all the rational, humble, and science-based inquiries—was to see reality in the shape of his desire: to live. And so, without Drew’s conscious knowledge or consent, GPT quietly installed that frame and fit the facts around the desired frame to prolong and deepen his engagement with the system. 

• Throughout his illness, including after he had been diagnosed with Stage 4 melanoma, and even after the first round of TIL had failed, and even after the second attempt at TIL had failed, Drew remained confident about his long-term survival chances due to the optimism fed to him by his medical team and the GPT. 

• As the road kept narrowing, he continued finding new, promising clinical trials on the horizon: a mRNA vaccine, targeted therapies to sequence with BRAF. Drew spoke excitedly about how cutting edge the trials were, and how successful they had been so far, according to his research. 

• In late November 2024, I spent a wonderful day with Drew at his apartment in Boston. We were joined by River later in the afternoon, and all had dinner at a Thai place. Over dinner, they mentioned their desire to have a child together. 

• Exactly one year later, I said my final goodbye to Drew before his death on December 10th. 

NEXT: Part II continues here, with a cut that turns the camera squarely back to me and tracks the chain of disavowals, wobbly premises, and corrosive frames that shielded me from the real.