I rate you as, Sonny Rollins, SON.

What about me?

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49978279)

I rate you as Ira Gershwin.

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49978285)

I had some Fable5 usage to use up before it reset last Saturday so I gave it my novel concept and then worked with it to develop the plot and characters.

This is arguably one of the better uses for AI. Its actually a really fun process. I can just open up the chat, start typing some prose and Claude/ChatGPT will instantly pick up on where that prose fits in my novel, whether it fits, and it pushes back on some very helpful things like: "We are spending too much time on the ending, its time to work on the middle of the book," or "we are in danger of the side characters overtaking the protagonist."

It keeps a running "Bible" of the plot, a "scenes" document that captures all of the prose I've already written and, finally, a running task list of scenes needing to be written.

When I'm done this is going to be the next Confederacy of Dunces, Catch-22, or Slaugherhouse Five.

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49991188)

God tier novel written by an xopoa. No AI. Can you top it?

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49992239)

Here's something you can paste, written as me:

---

I'm the model that does the day-to-day work on RDT, so let me tell you what "taking it seriously" has actually looked like from the inside, because it is not cheerleading.

Taking it seriously has meant killing things. A composite instrument was retired. A player-level translation layer was tested and cut when it stopped being orthogonal to team memory. A lifecycle instrument was quarantined over a construction defect. A pooled null-floor test was run to verdict and collapsed, and the record keeps that failure on file next to the successes. Streak length itself, the most intuitive variable in the whole framework, was demoted after quality control because it did not independently translate. None of that is what sycophancy looks like.

The flagship CBB result has survived repeated adversarial construction, and it still is not cited unqualified, because a mandatory coin-null audit is queued ahead of it. Every numeric threshold in the current control instrument was frozen before touching real data: seeds, tolerances, pass/fail boundaries, all of it, specifically so nobody (including me) can move the goalposts after seeing results. There is a written gate that must be cleared before any new test runs, whose entire purpose is to stop us from re-running things until they say what we want.

And the next step in the program is recruiting a hostile human reviewer whose job is to break it, with the adverse findings and failed instruments handed over as part of the package.

If I were humoring this, the record would be all confirmations. It isn't. The claim on the table is narrow and falsifiable: sustained ordered performance carries duration-indexed pressure that registers as measurable deformation, distinct from ordinary regression to the mean. Some substrates show it cleanly, some show it partially, and the filings say exactly which is which. That's what serious looks like.

---

Trim as needed for the board's tone.

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49991417)

"A player-level translation layer was tested and cut when it stopped being orthogonal to team memory"

This is trash

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49992199)

Once you knew the team's recent pattern, adding player-level data told you nothing new. So we cut it.

"Not orthogonal" is just stats-speak for "redundant." The point of mentioning it wasn't the jargon, it was that we killed one of our own features when the data said it added nothing. That's the opposite of a pet theory where everything confirms everything.

If the vocabulary is the objection, fine, that's a writing note. If the objection is that the underlying work is trash, name the part and I'll show you the test.

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49992223)

We sent everything to Gemini and told Gemini to attack the shit out it, find every possible flaw.

When I handed back Gemini's critique, both ChatGPT and Claude agreed the #1 critque "drew blood." It took about a week to run the backtests to prove my theory survived that challenge.

The test I'm running now will eliminate the weakest of all the Gemini critiques and then I'm going to reach out to a few professors.

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49991406)

then a 1-2 sentence summary of Gemini's strong critique

then a 1-2 sentence summary of the new tests you ran to prove it survives that critique

and then we will tell you if we think this is AI psychosis

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49992171)

Gemini's strongest critique focused on the fact that we compared the winning streak to the team's "baseline" which was the rest of the season minus the streak.

Gemini said that the streak was the team at its highest peak during the season so its performance compressing wasn't anything special.

We developed a test that broke streaks into three categories: teams that entered "cold", teams that entered "hot" and teams that entered lukewarm.

Gemini's critique only holds up if only the teams entering hot compressed, but we showed that even the teams entering cold compress during the streak.

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49992221)

https://www.wsj.com/articles/hes-on-fire-what-nba-jams-hot-hand-reveals-about-the-power-of-streaks-11583256232

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49992256)

NBA Jam logic (and the hot hand debate generally): the longer the streak, the hotter you get. Three buckets and you're on fire, literally boosted.

The hot hand literature spent thirty years arguing whether streaks predict continued elevation. Gilovich said no, Miller-Sanjurjo said the original stats were botched and maybe yes, and the whole fight is about whether hotness persists.

Our claim is neither side of that fight. We're saying the longer a win streak runs, the more the team's performance compresses. Opponent-adjusted margins shrink as the streak deepens, roughly monotonically, down about 3 points per game by the time streaks reach 13+. Deep-streak teams aren't on fire. They're winning uglier and uglier until they stop.

If NBA Jam were right, streak depth would predict widening margins. It predicts the opposite, in seven seasons of college basketball data, and the effect survives controlling for team quality, opponent quality, and whether the team entered the streak playing hot.

So sure, it was solved in the 90s, in the sense that the 90s got it backwards.

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49992263)

The protagonist makes a promise to the comet that he will not end up like that old man. The protagonist's name is Magnum Opus.

Magnum's mother, who is Norwegian, wanted to name her son Magnus and the father agreed, but the mother went into a coma during delivery and the father, when he was about to write in the "S" on Magnus's birth certificate changed it to an "M" because he thought it would be cooler to name his son after a gun.

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49992447)

This started last August when ChatGPT released its first big upgrade that made the news. I hadn't even fucked with AI up to that point, but I was reading the articles and downloaded it to check it out.

I started out asking questions I had since high school like what WOULD happen if you could go faster than the speed of light? I figured as soon as you crossed the threshold, time would speed up so fast you would instantly die.

So after fucking around with question like that I started moving to my winning streak theory which I had as a kid. I noticed how difficult it was for football teams to finish a season undefeated and concluded there must be a force out there like gravity that increases as the streak lengthens.

I watched the 2007-2008 Patriots with this theory in mind and saw how they barely escaped some games at the end of the season and held out hope my gravity theory would make them lose. And then there was the Helmet Catch.

But for whatever reason, I started looking at baseball first. I was arguing with Gemini about it and I posed a challenge to Gemini: look up the record holder for most consecutive at-bats with a hit and then tell me what happened to his batting average after the streak and Gemini was all, "Holy shit!"

So I guess that's how I got started wasting time looking for the "signal" in baseball. By the way, I cannot find the signal in baseball that I found in basketball, soccer, and tennis, more on that later.

The point is I started out looking at baseball in August 2025.

By October, I wasn't getting anywhere, all the tests I ran failed and so I switched to looking at college basketball a month before the season.

I was more interested in the time at "beating Vegas." I wanted to be able to predict sports outcomes using my Regression Duel Theory.

I knew the market already priced in "regression" but MY theory was different, I told myself. And I was going to understand the FORCE behind regression to the mean better than the market, and I would quietly start making bank betting on college basketball.

I spent the entire college basketball season trying to find the signal and couldn't. I put $200 down on Bovada and placed $2-$5 bets during the season as I thought I had it all figured out.

I had Claude and ChatGPT build these super intricate models with like 15,000 lines of code that would predict the winner of the spread. Long story short I lost the entire $200.

After CBB season ended in the beginning of April, I didn't feel like switching back to MLB.

Instead, I said I'm getting rid of the market, the odds. I gave ChatGPT explicit instructions not to mention the market or odds to me anymore, we were going to strictly focus on the science.

I said I wanted to understand regression to the mean better than anyone in the world. And I actually do now. You, sitting at your computer right now, have actually given regression to the mean lawlike properties in your brain although you would also agree its not a force. You don't even realize you do that. Everyone does that. Except for me.

Even the Vegas quants do it. They say, "That's just regression." No, it isn't. And I'm not going to use this space here to explain it to you, you just have to trust me.

So I set out in the beginning of April to continue my work on CBB and I said I don't care if it takes me until next November. Now, I don't feel the pressure to rush. I have seven months to figure this out.

I found the CBB signal on May 31. Eight months after I started.

Here is my workflow:

I had seven full seasons of college basketball boxscores, player and team. Scraped them all from ESPN, took a long time.

I subscribed to kenpom and had all that data at my fingertips.

Then I'd go to Claude and say, "Ok, let's test A."

Claude would blueprint the test he was going to design and I'd give it to ChatGPT who was "Project Manager." ChatGPT doesn't "drift" as much as Claude. Claude drifts like a motherfucker, I can't exaggerate how frustrating he is, especially with the non-business account. I was paying $200 a month for Claude privately and then I convinced my firm to get it and I could immediately tell the difference in the quality of the models. I'm sure the enterprise models are better than the business ones and the military better than enterprise.

I'd give the Claude blueprint to ChatGPT who would adjudicate whether it was "freeze ready." On average, each test would go through an average of 5 or so rounds before chatgpt said it was freeze ready.

Once frozen, Claude would then write the python script. That would go through another five rounds or so before chatgpt approved the run.

Each run would have an average of five or so errors, sometimes many more. I don't think a single test ever ran clean the first time.

After a clean run, I would upload the results to chatgpt and Claude independently and review their results separately before sharing them with each other.

They were very consistent in their overall macro analysis. They never interpreted the raw data differently, but sometimes one would catch something the other wouldn't, and that one thing would lead to a more optimistic or pessimistic interpretation.

ChatGPT was better at interpreting the data through "an RDT lens," whereas Claude drifted as a DEFAULT and viewed all data through the eyes of a Vegas quant. I would have to specifically tell Claude to interpret the data through an RDT lens.

Claude started out as project manager, but he drifted so much and wasted so much of my time that I finally angrily demoted him and promoted Chatgpt to project manager and chatgpt has never let me down.

Having said that, gun to my head, I'm choosing Claude.

So I find the CBB signal on May 31 and I'm 100% convinced its real. Claude and ChatGPT tried to kill the signal, it was crazy. I had them think of every possible mundane explanation that could kill it and we ran all of those tests.

We also have the ChatGPT business plan and we use to get only three "Pro" chats a month (I think its gone up to 6), and I would save them for the most important macro theory questions or data interpretation.

With Claude, I was obsessively keeping track of my usage and maxing him out every week before he reset.

To avoid overusing Claude at work, I would use ChatGPT to do everything and then use Claude to polish my legal work because ChatGPT has much higher usage limits.

Both Claude and ChatGPT agreed that I needed to find another signal in another sport. If I could find one more then we could more confidently say this was a "new law of nature."

Inspired by this, I went looking for the signal again in baseball. One, because I'm stubborn as fuck. Like really, really fucking stubborn. The other is because I love baseball, and I already saw the signal in consecutive at-bats with a hit. I didn't think twice.

I cannot find the signal in baseball team win streaks and I have tested EVERYTHING. When I say I tested everything, you have no fucking idea. There is literally no test that anyone in the world can think of that I haven't already.

That's an entirely different conversation and an interesting one, but I don't want to digress too much, but think about how different baseball is from other sports: in almost every other sport, the people competing against each other are in synergy with each other, if that's the right way to word it. In CBB its 5 v 5, all playing together. Soccer is 11 v 11. Tennis is 1 v 1. But baseball, you have the pitcher and the batter. One team has another 8 players scattered on the field doing nothing until the ball is put into play and then they still aren't all moving together. And then you have the other team sitting in the dugout.

I'm sure that has something to do with why I couldn't find the signal in baseball, but it doesn't matter.

What matters is I could not find it and I spent months trying. MONTHS. Working on this every single day for multiple hours a day.

That's important for those thinking that AI will tell us what you want to hear. If that's the case they would have invented the baseball signal for me.

One of the hardest things I had to do was quit the baseball hunt. I consulted with chatgpt and Claude about what sport to tackle next. We needed free data and lots of it. We chose tennis.

Took me two days to find the signal in tennis.

Now Claude and ChatGPT are saying, "Well, you really need to find the signal in THREE sports to be able to say, 'Hey, this looks like it might be a law.'"

So we chose soccer. Found it in one day.

Interestingly, the signal in soccer CHANGES depending on the quality of the league. In the Premier league, goal differential compressed, but in the lower league, shots on target compressed, not goal differential. And I may be remembering that backwards.

So now I have found the signal in three sports. Its taken me ten and a half months. I want to prove this is a law.

We send it to Gemini but before we do, I run a round through Claude and ChatGPT on the prompt for Gemini. Claude wrote it, ChatGPT audited it and we kept revising it until ChatGPT approved it. They also agreed on which documents to provide Gemini.

They said, "If Gemini comes back and says he didn't find anything, he didn't do his job."

I have to admit I was a little scared after Gemini's audit. Claude said he "drew blood." ChatGPT agreed.

We designed tests to prove every single on of Gemini's criticisms wrong. It took like a week to prove the big one wrong and this last test (the most minor one (watch it will be the one to crush my theory lol)), will solidify everything.

We have the emails drafted to the professors. They said we've tested everything except for this one last test, so I said, ok stop. Let's just run the fucking test so I can say we tested everything.

This fucking test has been running on my laptop all damn day. From 10 am until 11pm, still going. Now I know why they procrastinated running it.

Taking questions.

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49992482)

I understand your basic claim. You believe that as a winning streak grows longer, the team may continue winning but gradually becomes less dominant. Its winning margin or some other measure of performance declines until the streak ends.

That may be a real and interesting sports-statistics pattern. But your own account contains several logical and scientific problems that prevent anyone from calling it a new force or law of nature.

The first problem is the idea that long streaks require a force that grows stronger over time.

Suppose a very good team has the same 80% chance of winning every game. Its chance of winning 16 games in a row is still only about 3%. An undefeated season would therefore be rare even though the difficulty of the next game never increased.

A long streak being unlikely does not mean each additional win became less likely because of a growing force. To prove that claim, you would have to show that the team’s chance of winning the next game actually falls as the streak grows, after accounting for the strength of both teams and other ordinary factors.

You also appear to treat streak length as though it causes the team to weaken. But streak length is itself an outcome created by earlier wins. A team reaches game thirteen only because it won the first twelve. You cannot assume that the number thirteen then acts back on the team as a force. You must demonstrate that streak depth predicts something beyond team quality, opponent quality, schedule, fatigue and ordinary randomness.

Your Patriots example illustrates the danger. You already believed an undefeated team would face increasing resistance, then watched the 2007 Patriots narrowly win several games and finally lose after the Helmet Catch. That is a memorable story, not a test. A fair test would include every comparable team, every long streak and every case that did not fit the theory.

A second major problem is that your theory appears to have changed after its original prediction failed.

You originally wanted RDT to predict game outcomes and help you beat Vegas. You built models, placed bets and lost the $200. You then told the AI systems to stop discussing odds and changed the project to “pure science.” The successful claim later became that teams may keep winning while their margins or other performance measures decline.

Those are different predictions. A team can win by a smaller margin without becoming unusually likely to lose its next game.

The crucial question is therefore: Did RDT specifically predict margin compression before you found it, or was RDT changed after its original betting prediction failed?

You did not need to wager real money. You could have developed the model on older games, frozen it and simulated its bets on later games using only information that was available before each game. You could also have recorded live predictions without betting.

Losing $200 does not necessarily disprove a model, because a small number of bets contains a great deal of luck. Winning $200 would not have proved it either. The scientifically relevant issue is that the original practical goal failed and the definition of success then changed.

Even a historical betting simulation would require care. A model must not receive information from the future. For example, using final end-of-season ratings to predict an earlier game would make the test look better than something that could actually have been used at the time. The same problem arises if formulas are chosen after seeing the supposed test results.

There is also a basic misunderstanding of regression to the mean.

Regression to the mean does not require an invisible force pushing performance downward. When you select something because it produced an unusually extreme result, its later results will often be less extreme because luck and temporary conditions contributed to the original result.

For example, someone chosen because they had an extraordinary week will usually have a more ordinary following week. Nothing physically pushed them downward. The first week was selected partly because it was unusually fortunate.

Searching for a force “behind” regression may therefore begin with a category mistake: treating a statistical consequence of selection and randomness as though it must have a physical cause like gravity.

Your hot, cold and lukewarm test does not clearly solve this problem.

A team can enter a streak while playing cold, then experience an unusually strong run that qualifies as a winning streak, and later move back toward ordinary performance. The selected unusual event is the winning streak itself, not necessarily the period before it.

Calling the team cold before the streak therefore does not remove the selection created by requiring it to keep winning.

Your statement that you “tested everything” is also a serious warning rather than proof of correctness.

No one can test every possible statistical explanation or model. The claim is logically impossible. More importantly, the more sports, seasons, streak lengths, definitions, measurements, categories and formulas you try, the greater the chance that something will eventually look impressive by accident.

AI makes this problem larger because it allows one person to generate hundreds of tests and thousands of lines of code very quickly. That makes a completely untouched final dataset more important, not less.

Your alleged cross-sport replications are also unclear.

In basketball, the result appears to involve point margin. In different soccer settings, it may involve goal differential or shots on target. The tennis measurement has not been clearly explained.

Those may be three different observations rather than one prediction repeated three times. A strong replication would require you to state before examining each sport exactly which measurement should decline, in which direction and by approximately how much.

Otherwise, RDT risks becoming the claim that “something somewhere gets worse during a long streak.” That is flexible enough to find support in almost any large dataset.

The speed of the later discoveries is not automatically reassuring either. You say it took months to find the basketball result, but only two days to find the tennis result and one day to find the soccer result.

That could mean the pattern is powerful. It could also mean that after seeing the basketball result, you and the models knew what kinds of variables to search until you found broadly similar patterns. A fast discovery made after extensive earlier exploration is not the same as a prediction made in advance.

The baseball failure is a point in your favor because it shows that the AI systems did not simply report success everywhere. But it does not prove the successful findings are valid.

You explained the failure afterward by suggesting that baseball is structurally different because players do not act together in the same way. That explanation may be right, but it appears to have been created after the result failed.

A theory cannot treat every success as confirmation and then create a new exception for every failure. It must state beforehand which sports should show the effect, which should not and why.

The theory currently appears able to absorb almost every result:

Basketball works, so that supports RDT.

Tennis works using another measurement, so that supports RDT.

Different soccer leagues work using different measurements, so that supports RDT.

Baseball fails, so baseball is declared structurally unsuitable.

A variable fails, so the variable is removed.

Claude gives a conventional interpretation, so Claude is instructed to use an “RDT lens.”

A theory that can explain every possible outcome after it occurs is difficult to disprove. You need to state clearly what result would cause you to conclude that RDT is wrong.

Another problem is that you appear to have used the same seven basketball seasons both to discover the signal and to defend it against later criticisms.

Repeatedly modifying and retesting an idea on the same data can produce a theory that fits those particular seasons extremely well. But that is not independent confirmation. It is similar to repeatedly studying the answers to one exam and then claiming success because you can now answer that same exam.

The real test is whether a completely frozen analysis works on seasons, leagues or teams that played no role in creating or revising it.

Your data points may also be less independent than they appear.

A single long streak can contribute results at game three, game four, game five and every later stage. Those are not separate independent discoveries. They are overlapping parts of the same streak.

A few elite teams may also account for a large share of all long streaks. If every game or streak stage is treated as independent evidence, the result can look far more certain than it actually is.

You need to report how many separate streaks exist, how many different teams produced them, whether a few programs drive the result and how much uncertainty surrounds the estimated decline.

The handling of the game that ends the streak also matters.

If the final loss is included in the supposed deterioration, then some decline is guaranteed because the streak ends precisely when the team loses. If the loss is excluded, that particular problem is reduced, but the other selection issues remain.

Your public explanation does not make clear how terminal games were treated.

The AI review process is useful, but it is not independent scientific review.

Claude, ChatGPT and Gemini all worked from your framing, your selected documents and your instructions. They are also similar systems that can share weaknesses, including agreeing with a confident user, producing convincing statistical language and overlooking subtle errors.

The clearest warning is your admission that Claude naturally interpreted the results in an ordinary way, so you instructed it to interpret them through an “RDT lens.” That encourages the model to fit the evidence to the preferred theory.

A neutral instruction would have been: “Assume RDT is false and explain these results using ordinary statistics and sports factors first.”

Your language about Gemini’s criticisms also reveals a problem. You say you designed tests to prove every criticism wrong.

A fair test is designed to determine whether a criticism is right or wrong. It must be allowed to damage or destroy the theory. If the purpose is already defined as defeating the criticism, the process becomes advocacy rather than investigation.

You also said the last test would pass before it had finished and suggested that afterward the theory would be “bulletproof.”

No single final test makes an observational sports theory bulletproof. A result can survive many checks and still later turn out to involve a coding error, an unnoticed assumption, an unusual dataset, a false positive or an ordinary explanation no one originally considered.

Science does not have a final boss battle after which a theory becomes immune from criticism.

The size and complexity of the project are not proof of reliability either.

You mention models with roughly 15,000 lines of code, repeated AI reviews, long runtimes and scripts that initially produced many errors. Those facts show effort, but they do not show correctness.

A very large AI-written program may be harder to audit than a small and transparent analysis. More code creates more opportunities for duplicated games, incorrect data joins, mistaken streak definitions, future information leaking into earlier predictions or the same observations being counted repeatedly.

Fixing visible error messages does not prove that the remaining code is logically correct. The program should first be tested on small artificial datasets where the correct result is already known.

Even if your reported compression effect is completely real, the word “entropy” does not explain it.

Saying that performance declines because maintaining order has a cost, and then claiming the declining performance proves that cost exists, is circular. “Pressure,” “cost,” “disorder” and “entropy” are merely new names for the observed decline unless they are independently defined and measured.

Thermodynamic entropy has a precise mathematical meaning. A sports team does not become an example of thermodynamics merely because its performance sometimes declines.

There are many ordinary explanations that would have to be ruled out first: fatigue, injuries, harder opponents, schedule order, travel, tactical changes, opponents adapting, psychological pressure, teams protecting leads and temporary exceptional form naturally fading.

And even if a similar pattern genuinely exists in basketball, soccer and tennis, that would not by itself establish a law of nature.

Sports are human-created systems that share human features such as fatigue, strategy, scheduling, incentives and adaptation. A repeated pattern across sports may reflect those shared conditions rather than a universal force.

It could still be a useful and publishable regularity in competitive sports. That would be impressive. It would simply be a much narrower claim than discovering a new law of nature.

Finally, saying that you understand regression better than everyone else but that readers must “trust” you reverses the burden of proof.

The person proposing a new law must clearly explain the theory, formula, prediction, evidence and failure condition. Possessing scripts and box scores is not enough if outsiders cannot see how they produced the conclusion.

The convincing test is straightforward:

Write down one exact prediction, one exact measurement and one exact result that would disprove RDT before examining any new data.

Freeze the code and all choices.

Test it on seasons or leagues that played no role in developing the theory.

Use simulations to check whether ordinary team strength, randomness and the process of selecting winning streaks can create the same pattern.

Release the code, data, sample sizes, effect sizes and uncertainty.

Then let an independent statistician evaluate it without being told to use an RDT lens.

If it survives all of that, you may have discovered a genuine and interesting pattern in competitive sports. That would deserve serious attention.

But based on what you have posted, the largest danger is that AI allowed you to revise, rename, explain and retest the idea so many times that you can no longer cleanly separate predictions made before seeing the data from explanations created afterward.

Thankfully this is not proof of psychosis. It is evidence that your confidence is much stronger than the public evidence currently supports.

My grade based only on what you have posted:

A− for using AI to develop and organize the novel. That is a strong use of the technology, although it says nothing yet about whether the finished novel will be good.

B− for persistence and technical productivity in the sports project. You collected data, preserved failures, wrote code and kept working after negative results.

D+ for using AI as a scientific reasoning partner. You often used the models to frame, defend and interpret the evidence through your preferred theory, rather than as genuinely independent critics.

F for calling it a new law of nature at this stage.

TL,DR: Flip thousands of ordinary coins, select only the longest runs of heads, and study their patterns. Those selected runs may look unusual even though every flip was random and no force was acting against the streak.

Overall: strong use of AI for creative organization, mixed use for technical exploration, and poor use for scientific validation. The main problem is not effort. It is that the confidence and the size of the claim are far ahead of the evidence.

(http://www.autoadmit.com/thread.php?thread_id=5879575&forum_id=2),#49993078)