A free preview, generated by MasteryLoopHQ.
A viral video in late October 2026 makes a bold claim. A person picks a picture and links it, in their mind, to a reference number. They give an AI chatbot only the number and ask it to describe the picture. The video says this works every time. In the video's own example, the chatbot was given "target id# 850." It replied with what it called a "speculative guess": "a light-colored building with a row of windows or columns, beside greenery." The picture was a pyramid-shaped building with light, vertically ribbed siding and four small dark windows, under a blue sky with trees at the edges. The chatbot's guess mentioned "A broad rectangular structure," and its sketch showed a long, flat-roofed building. A reaction post to the video told a similar story. Another person pictured a beach, and the chatbot described a wide, open coastal area with a path, thin poles, low dunes, water and a hazy sky. They felt it worked. When a description lands near the picture in your head, it is startling, and that feeling is real. The rest of this article asks a different question: how would we know it is more than chance?
Remote viewing is the idea that a person can describe something hidden from them without using their ordinary senses. A declassified US government briefing defines it this way: "REMOTE VIEWING - The Acquisition and Description, by Mental Means, of Information Blocked from Ordinary Perception by Distance, Shield, or Time." In the version the video shows, one person picks a target and links it to a number, and a viewer tries to describe the target without knowing what it is. The US government ran and funded remote viewing research from the early 1970s into the mid-1990s, ending after a 1995 outside evaluation. This article tests only one claim: that someone can describe a hidden picture better than chance.
Psychologists have names for what can happen here. The American Psychological Association describes the Barnum effect as "the tendency to believe that vague predictions or general personality descriptions ... have specific applications to oneself." In plain words, a vague description can feel like it was written just for you. The APA credits Bertram Forer, "who first studied it in 1949." A related idea is confirmation bias: "the tendency to gather evidence that confirms preexisting expectations." In plain words, we notice what fits and skip what doesn't. Those ideas come from research on personality descriptions and expectations. Ray Hyman, a reviewer in a 1995 evaluation of remote viewing, described a similar problem with judging matches: "subjective validation comes about when a person evaluates the similarity between a relatively rich verbal description and an actual target or situation. Inevitably, many matches will be found." A list like outdoor, bright daylight, greenery and straight lines fits a huge share of photos. A list like coastal, water, sky, path and low plants fits most beach photos. In the video, the person who knows the target is also the one judging the match. So they notice the parts that fit and pass over the parts that don't. Now take the video's own example. The guess said "A broad rectangular structure" with "Straight lines and a strong horizontal arrangement." The sketch showed a long, flat-roofed building. The target was a pyramid-shaped building with vertically ribbed siding. What they share is daylight, a light-colored building, a few windows and trees. Many photos share those features. A 1995 evaluation of remote viewing by the American Institutes for Research said something similar about real operational reports: "The reports were most likely to prove accurate with regard to general stereotypical characteristics of the situation." The report added that such results "might be attributable to the background information available to the viewers." It also said: "The information provided in the reports was stated in broad, vague terms."
A chatbot only has the words you give it. In the video, the prompt was "Remote view target id# 850." Words like "remote view" and "target" may point the chatbot toward a picture of a place or an object. The chatbot also labeled its own reply a "speculative guess," and the drawing was labeled a "speculative sketch." The reply says the guess was "fixed before the reveal," but nothing on screen shows when it was written. One simple explanation worth testing is that when a chatbot gets only a number, a typical scene (daylight, a building or a landscape, some greenery) is an easy reply to write. This article does not claim to know how any particular chatbot works inside. A fair test can check the idea. We also see only the run the video chose to show. The video says several experiments were run, but it does not show how many, or what the others gave. The second demo's prompt is not shown either, so we can't tell what the chatbot was given. Nothing on screen shows earlier chat history, memory or custom instructions either.
Before you are impressed by a hit, ask how often a description like that would match any picture of that kind. Here is a way to try it. Write one generic description, such as "outdoors, daylight, a structure, some plants, sky." Hold it up against ten random vacation photos and count how many it fits. Then think back to the examples. Both examples used common subjects, a building and a beach, which a general description can fit easily.
The video connects its result to quantum physics. Entanglement is real physics. The Nobel Prize website explains: "What happens to one particle in an entangled pair determines what happens to the other, even if they are really too far apart to affect each other." The 2022 physics prize went to Aspect, Clauser and Zeilinger "for experiments with entangled photons, establishing the violation of Bell inequalities and pioneering quantum information science." But entanglement does not let anyone send a message. The Caltech Science Exchange says: "A common misconception about entanglement is that the particles are communicating with each other faster than the speed of light... Experiments have shown that this is not true, nor can quantum physics be used to send faster-than-light communications." Caltech professor Thomas Vidick puts it simply: "There can be correlation without communication." Even quantum teleportation needs an ordinary message. The Stanford Encyclopedia of Philosophy says: "Bob can reconstruct the state ... on the basis of just two bits of classical information communicated by Alice." Physicists describe the linked results as compatible with a "no-signalling condition." In plain words, whatever one side chooses to measure, the odds of the other side's results stay the same, so neither side can send a message this way. The link shows up only when the two sets of results are compared, and comparing them needs an ordinary channel. None of these sources says entanglement links a thought to a number, or a person to a chatbot. A result we can't yet explain is a gap in an explanation, not evidence for a quantum cause. This article does not try to settle bigger questions, such as whether we live in a simulation. These sources do not address them.
A 1995 report by the American Institutes for Research evaluated the government's remote viewing research. In it, reviewer Jessica Utts described a fair way to judge a result. A judge who does not know the true target is shown the response and five possible targets. One is the right answer, and four are "decoys." Each of the five "must have had an equal chance of being selected as the actual target." The report adds: "The probability of a first place match by chance would still be only one in five." Here is what the report found, stated fairly. Its evaluators wrote that both reviewers, Jessica Utts and Ray Hyman, noted that "a statistically significant effect has been obtained" in the laboratory studies. They disagreed about the cause. Utts concluded that "psychic functioning has been well established." Hyman tended to agree that "real effects are occurring in these experiments," but wrote that "the occurrence of statistical effects does not warrant the conclusion that psychic functioning has been demonstrated." The evaluators reached their own conclusion: "at this juncture it would be premature to assume that we have a convincing demonstration of a paranormal phenomenon." On real intelligence work, they wrote: "Remote viewing failed to produce actionable intelligence." Edwin C. May, who says he directed the research for 10 of its 24 years and so is an interested party, answered in a 1996 commentary that the review looked at only a small, late sample of operations. Both reviewers also agreed on one flaw: "only one judge—apparently the Principal Investigator—was used in assessing matches throughout these experimental studies." With a single judge, no second judge checked whether the matches held up. In a home test, if the person who picked the target also does the judging, that judge is not blind at all.
Here is a test you can run. (1) Person A, who picks the target, chooses five photos that look very different from each other, such as a city street, a forest, a kitchen, a beach and a stadium. Very different photos are our design choice, not a rule from the report. Build the whole set before anyone asks the chatbot anything. (2) A random pick, not a choice, selects the target. Roll a six-sided die and roll again if it lands on 6. Person A writes down a reference number, seals the answer, and then has no contact with B or C until judging is done. (3) Person B, the runner, has not seen the photos and does not know which one is the target. B gives the chatbot only the number and saves the full reply with the time. Use a fresh chat each time, with memory turned off. (4) Person C, the judge, has not seen which photo is the target. C gets the reply and the five photos in shuffled order, with neutral file names, and ranks all five photos from best match to worst match. (5) Unseal the answer and record whether the target was ranked first. (6) Repeat many times with new photo sets, and record every trial, not just the impressive ones. Before you start, write down how many trials you will run, how you will score them, and any rule for dropping a trial. (7) Compare to chance. With five photos, a random guess ranks the target first about one time in five. In 20 trials, chance alone gives about four first-place hits on average, and swings of a couple of hits either way are normal. To beat chance at the usual 5 percent level (one-sided), you would need 8 or more first-place hits in 20 trials. A small real effect could still hide in 20 trials: Utts notes it can take "hundreds or sometimes thousands of trials." Watch for common ways a test leaks: someone who knows the target types the prompt, re-prompts or picks among replies; the judge knows the target; the prompt hints at the category; misses get dropped; the target is not picked at random from the set; the scoring or dropping rules change after the results are in; or there are too few trials. Researchers have also preregistered an exploratory test of this very idea with an AI chatbot (40 trials, five-image sets). As of October 26 the registry lists no results.
A recent viral video claimed an AI chatbot can describe a picture that is only linked to a reference number in someone's head. The descriptions felt like hits, but they were broad scenes (daylight, a building, greenery, a path, water, sky) of the kind that fit many photos. In the video, the person who chose the target also judged the match, and we see only the run the video chose to show. Entanglement is real, but as a Caltech professor puts it, there can be "correlation without communication," not a way to send information. The US government ran remote viewing research from the early 1970s into the mid-1990s. Its 1995 evaluation found a statistically significant laboratory effect that the two reviewers read differently, said the operational reports were "stated in broad, vague terms," found that its information had never been used to guide intelligence operations, and concluded that it would be premature to assume a convincing demonstration of a paranormal phenomenon. Edwin C. May, who says he directed the research for 10 of its 24 years, later disputed the operational finding. A fair test is simple: a blind judge, decoy photos, a one-in-five chance baseline, every trial counted, and enough trials to mean something.
In your own words: (1) Why does a description like "outdoors, daylight, some greenery" feel like a hit, and why doesn't that show it saw the target? (2) What can entanglement do, and what can't it do? (3) Design a fair test: who knows what, how many photos, and what score would chance give you?
Reading a good explanation feels like understanding it. Usually it isn't the same thing — and you don't find out which one you've got until someone asks you to explain it back.