Monday, April 14, 2025

An Antidote to AI Intellectual Property Theft

Artificial Intelligence software technology has generated tremendous press over the past three years as multiple companies released competing systems with lofty claims about the capabilities of those systems. Those same releases have scared the bejeezus out of currently well-paid "think workers" (programmers, musicians, graphical artists, knowledge workers) by suggesting their work can be / will be replaced by AI based systems within MONTHS. These AI stories have also generated enormous attention on a related topic – how the firms creating these AI systems "trained" the expertise into those systems by pointing them at PETABYTES of text / image / audio / video content created by other people. This approach to training is legally problematic because none of these firms obtained prior consent from even a fraction of the creators whose work was used for training. And these systems are not just for-profit systems that won't be sharing those profits with the creators, they are for-profit systems being promoted as REPLACEMENTS for the work those creators perform, posing a direct threat to their future work and income.

This highly illegal, highly automated, wholesale violation of intellectual property rights of millions of individuals poses a variety of technical challenges to the larger economy. However, software developers familiar with the operation of AI systems in both training mode and "production" mode are fighting back. New tools have been developed by creators who know AI operators are feverishly attempting to negate potential copyright infringement rulings in the courts both retroactively for their prior theft and going forward to allow it to continue These new techniques might be enough to thwart the exponential growth of AI on jobs in the near term. They might also reflect a huge financial risk for those who invested heavily in AI software firms and the hardware firms making billions selling the computers, storage and networking to support gigantic AI deployments.

Before explaining how these "AI antidotes" work, a short explanation of how AI systems "learn" and operate is required to explain a core flaw in the resulting system that poses problems for operators and intellectual property owners alike.


How AI Systems "Learn" (Simplified)

Most of the AI models referenced in the public today use either a Large Language Model architecture that accepts text as input and generates text as output or Convolutional Neural Network models optimized for analyzing image data for object recognition or new image generation. Generically speaking, these architectures take a vast collection of inputs provided to "train" the system and chop that data into binary data representing small pieces of the total source content. These sub-tokens are read into memory and millions of mathematical operations are performed on collections of those tokens analyzing the statistical probability of token Z appearing after tokens X and Y. If the training is aimed at creating a model for English prose, content in the training data might contain a string of text like this:

And she's buying the stairway to heaven

Of course, many English speaking humans might immediately recognize that as a lyric within a song. But the AI "knows" nothing. It doesn't even see that as a sentence or phrase. That text might be tokenized down to 1-2 character chunks so it is actually processing THIS:

An d sh e's buy in g th e st ai rway to he av en

(And remember spaces are just another token to this logic like any other character.) But the training process has scanned TERABYTES of similar English prose, not just one line. As those TERABYTES were tokenized then analyzed, the resulting statistics would provide tables that reflect the fact that in the training data,

  • the combination SH is usually followed by a vowel or space
  • the combination SHE is often followed by either a space or apostrophe
  • the combination SHE' is nearly always followed by an S
  • the combination TH is often followed by a vowel with E being most common

The training process refines those statistics by periodically scanning a separate cache of data NOT included in the training data then comparing how the predicted "next letter" from the first generation of statistics matched the ACTUAL "next letter" in some of the test data. Where the first generation statistics are found to be significantly off, the training algorithm re-weights those statistics, another round of training is done and the process is repeated. This process generates ADDITIONAL statistics that show how much closer predictions match test data in each round of training to track diminishing returns of spending more compute time on training. When the prediction rate stops improving, training is deemed complete.

So with tens of thousands of probabilities like this derived from analyzing PETABYTES of examples of English prose, accessing those probability tables with an "agent" user interface that accepts a "prompt" of text from a user makes it possible to generate text via statistics that seems like a response to the human's prompt.

What about audio content? With audio data, the low-level "tokens" processed by the training algorithm won't be ASCII codes for characters, they will be some small sample of the binary data of the audio waveform, maybe a half millisecond. If you imagine Robert Plant's voice – or ANYONE's voice – saying or singing that lyric, the vowel and consonant sounds will have characteristic waveforms which result in characteristic digital data patterns that will eventually coalesce, allowing the pure audio input data to be mapped to the lyric text.

What about images and video? The same process applies. Imagine a green square in the middle of a white background as reflected in a bitmap image file. When used to train an AI, the AI can identify the width and height of the overall image file in pixels, then find patterns that the RGB color coding of pixels flips between green and white and the (x,y) pixels where the color changes have some predictable pattern to them. Without anyone ever writing an actual program to TELL the AI what (x,y) scheme would reflect a circle versus square versus rectangle versus triangle, the AI can derive that mapping by scanning millions of images, generating statistics, then running an audit

This two-stage training and processing architecture exhibits two key flaws, one that affects both creators and AI operators and another that affects AI operators and their end users. The first problem will be termed the "provenance problem" and the second problem will be termed the "hacking problem."


AI's Provenance Problem

The provenance problem stems from the fact that an AI system's intermediate statistics optimized during training and its final "operational" statistics used by users do not retain metadata about the sources that drive any particular "inference" the system generates as an output. Furthermore, the tokenized nature of the statistics prevents any human-meaningful "search" from being executed to ask the system "where did you learn this?"

In the simplest terms possible, this loss of provenance information is exactly analogous to the following sequence of events:

  • someone providing you ten billion individual temperature readings for the globe
  • calculating the average of all ten billion measurements
  • taking the average value and transferring that to another system then THROWING AWAY all of the source data
  • then being told all of the values originating from source X (ten percent of the entire input data) were wrong and need to be adjusted or removed from the average
  • then being told the original raw data points and their statistics (number of samples, average value) cannot be re-supplied so they can be deducted from the original average

If you still had the original data points mapped to their source, these flawed inputs could be removed to allow the average to be recomputed with corrected data or without the flawed data. Without the original data, there is no way to "back out" that bad input from the resulting statistics. The only options are to accept the final model with those known flaws or retrain the model after attempting to block that source from being ingested again for training.

For creators of content, this provenance problem makes it virtually impossible to PROVE an AI operator ingested their content. There's no administrative tool a court can force the operator to run that will scan the training statistics or the operational statistics and find a specific probability driven by ingesting a specific source document. This gives operators of the AI system a huge margin of error in terms of plausible deniability that they DID knowingly scan creator X's content and incorporate that content into their model without permission. This is especially true because other middleman firms may have cached the content and the fetch of that content may have never hit the original creator's system to provide a hint the content was pulled.

For the operators of AI systems, this provenance problem is equally problematic because accessing the PETABYTES of online content needed is very difficult to automate efficiently if thousands of exclusionary rules must be analyzed prior to downloading content and feeding it into training cycles. And if unauthorized content IS injected into a training cycle, even if the operator sees proof after the fact their training DID ingest unauthorized material, there is NO METHOD for surgically removing that offending "knowledge" from the final system. This means AI system developers and operators potentially face huge legal problems if courts rule AI training IS violating copyright protections of individuals en masse.


AI's Hacking Problem

AI's provanence problem leads directly to AI's hacking problem. The need for AI operators to feed TERABYTES of data into training processes and the operator's unwillingness to devote appropriate labor resources to carefully curate content sources both for owner consent and appropriateness of the content leads to AI's hacking problem. It's the old "garbage in, garbage out" problem on steroids. It bears repeating. AI systems don't "know" ANYTHING. All they are doing is generating statistics based on consecutive sequences of very small units of data presented in their training data. Training algorithms are not analyzing the content being ingested at human contextual levels. This means ANY data present in the content fed to the training processes will be used for training.

This poses profound risks to operators and users of AI systems. If an operator trains a system and ingests a gigabyte of suspect content amid a terabyte, that suspect content may have data within it having nothing to do with the intended use of the AI system. If the goal is to create an AI model for optimizing derivative trades on corporate bonds but the AI training data included thousands of blogs pitching get-rich-quick schemes generated by robots for a bunch of crypto-coin fanatics living in their parents' basement, the "noise" from those bad sources will be present in that AI and alter outputs in unpredictable ways.

As one example, AIs for speech recognition might be designed to recognize "Hey Google, do X" and pass a command into the user's phone to do X. However, that AI process might detect a "Hey Google, do X" command in data inputs not perceived by the user. The user might be watching a video that embeds the same command amid other sounds that mask it from being detected by the human but can be spotted by the AI which then dutifully executes the command. That command might be "Hey Google, unlock the front door" which may leave your house wide open to burglars.


An Antidote for AI Intellectual Property Theft

The launch of these for-profit AI systems by giant corporations with stolen content has pissed off a lot of content creators. Unfortunately for the giant AI firms, those firms do not have a lock on all AI talent and software development expertise. And many people with AI expertise and software development skills are content creators who understand the threat posed by AI systems to their own livelihoods. And they have devised new tools to combat the wholesale theft of intellectual property by leveraging these two flaws AGAINST AI systems.

Remember, those two key flaws are:

  1. The inability of training processes to discriminate "good" and "bad" inputs (whether due to accuracy or security or legality)
  2. The forward-only nature of AI statistics during training and operations – the statistical impacts of "bad" data cannot be surgically identified, quarantined or removed from a model without starting over and explicitly excluding the "bad" input which may prove impossible to do.

Simply put, the antidote to this wholesale content theft involves altering content being posted on public sites with embedded data that doesn't interfere with HUMAN consumption of the content but CORRUPTS the larger data stream seen by a webcrawl engine and fed into AI training. It's the equivalent of having content appearing like this to a human:

And she's buying the stairway to heaven

appear like this electronically to the AI training process:

And encabulator she's turbo buying prefabulated the
aluminite stairway hydrocoptic to marzlevanes heaven

This "poisoning" of the content poses multiple problems to the AI operator.

  1. While masked from affecting a human user's experience, the AI will see and process ALL of the data since current ingest automation just slurps in the entire data stream.
  2. Altering training bot automation to detect and mask this bogus data requires parsing the raw file just like a browser, incurring additional processing times that would increase compute costs for training by 10x or 100x.
  3. Because of the provenance problem, once this polluted data enters the training realm, the AI operator has no viable way to back the data out.
  4. Because of the provenance problem, once this polluted data enters the training realm, the AI operator has no way to identify the source that actually supplied the data to make explicit efforts to NOT craw that site.
  5. Because the AI operator has no direct clue about how this poisoned data affects the model being generated, the AI operator potentially faces liability issues for harmful or inappropriate outputs it cannot predict.

How effective is this poisoning strategy? The 10:00 point in this highly recommended video by Benn Jordan

depicts what happens to the iterative learning curve when training ingests poisoned data. The rate of improvement bottoms out almost immediately in the training cycles. Bottoming out doesn't mean it reaches the desired quality level, it just means learning stops improving, making additiional training pointless at a much lower quality level. The video includes some samples of output from music generating AIs without poisoned data and with.

In other words, this technique has the potential to completely BREAK the business model for operating an AI off stolen data. Of course, there's another obvious implication of this "white hat" hacking technique. The very mechanism being leveraged by creators simply trying to protect their content can be used by bad actors to subliminally link insecure code or content into AI results by injecting similar hidden noise inside other content being scanned by AI operators.

And that's the real takeaway from this latest news in the world of AI development. There are unique flaws to the underlying mathematics and the financial business model of training and running AI systems that have already put the technology in the same technological circle of hell as the perpetual struggle between hackers and computer users and anti-virus software developers. The nature of the technology guarantees neither party can ever secure a permanent upper hand, putting an upper bound on the true value of the process. It will never get better, only different.


WTH

Thursday, April 10, 2025

The Crash That Wasn’t (Yet...)

News stories emerging after the last minute “pause” of draconian tariffs on virtually every good entering the US are converging on a common set of truths about what triggered the pause and the rebound in stocks. Or more accurately, news stories are converging on a set of truths about what didn’t trigger the pause.

It wasn’t due to Trump suddenly understanding the physics and mathematics of trade and tariffs.

It wasn’t due to Republican politicians threatening to join a vote explicitly negating Presidential declarations of a state of emergency used to legally rationalize the imposition of tariffs.

It wasn’t due to Republicans privately revolting after hearing vile comments from Trump at a Republican Party dinner the evening of April 8, 2025 bragging about how every country was “kissing his ass” in attempts to undo the looming damage.

So what DID trigger the pause?

On the surface, most of the coverage references unrest in bond markets. Under the covers, the real trigger involved overnight lending, a fundamental aspect of the modern worldwide banking system that usually operates flawlessly with as much drama as the oil pump in your car engine.

Most Americans have as much interest in understanding overnight lending in the financial world as they do learning about how the oil pump works in their car but in the current situation facing the United States, it is crucial that more Americans DO understand this process. Hiccups in this process pose existential threats to every American, not just to their retirement account balances but their daily economic livelihood. And the operation of this process is at risk every minute of the day that the current Trump and Republican cabal is in control of the American government.

The thesis boils down to this. Bad things in the economy don’t happen when investors lose money. Investors lose money every day. Bad things -- REALLY bad things -- happen when money stops flowing. Money stops flowing when parties to a proposed transaction feel unable to determine the level of risk they are taking in entering a deal with each other. That risk stems from multiple factors:

  • the ability of the party paying money to deliver the promised amount
  • the ability of the party receiving money to deliver the promised good or service being purchased
  • concerns of either party of the actual stability of the currency used as the medium of exchange
  • concerns of either party regarding larger systemic risks in the economy that might make the transaction impossible to complete or entirely moot

Regulatory processes applied to banks require a daily settlement of their books to confirm compliance with accounting regulations. These daily settlements generate an enormous volume of financial transactions between banks. These transactions are initiated by people who are presumably the most comfortable handling large transactions under stressful conditions that are changing rapidly and represent great risk. Statistics are collected from those trades to actually influence the transactions themselves (via interest rates) but those statistics also lend themselves to outside review that can divulge worries of those executing these transactions that might escape coverage in most media outlets.

When bankers at the core of the financial industry who should be most adept at evaluating financial risks cannot reach consensus, required lending at the core of the system can instantly seize up, triggering much larger failures far beyond stock markets. Understanding this process is crucial to understanding the warning signs currently being flashed to the larger economy. Understanding this process requires background in five key areas:

  • basic banking accounting terminology
  • basic bond behavior
  • fractional reserve banking
  • overnight lending
  • overnight rates

It’s not the most entertaining reading but the insight from this might be the only basis for building support for changes to prevent future failures.


Accounting Terminology in Banking

Most people know banks categorize the numbers representing customer accounts in terms of assets and liabilities. However seeing these terms in the context of banking institutions is often confusing because banks sit at the opposite end of the proverbial table from individuals. This reverses the assignment of labels to different financial transactions from what consumers are used to seeing. For an individual, $10,000 in a savings account is an ASSET, while a $40,000 car loan and a revolving balance of $7500 on a credit card are LIABILITIES. From a bank’s position, those labels are flipped. Savings accounts and checking account balances are LIABILITIES to the bank because a customer could initiate a transaction at any time requiring the bank to make those funds available via a cash withdrawal, paying a check or handling a debit payment. To the bank, a loan or a credit card balance is an ASSET because those amounts represent future payments they expect to collect from the customer.


Basic Bond Behavior

Bonds are financial instruments that provide a more efficient means of borrowing to large organizations that eliminate a direct, permanent tie from a specific borrower to a specific lender. When a consumer needs to borrow $50,000 for a new car, the consumer goes to a single bank, provides credit information and reviews the interest rate and repayment terms offered by the bank. If the consumer signs the loan, the bank hands over $50,000 NOW in exchange for the consumer making a recurring set of payments over X months which add up to $50,000 plus an additional amount reflecting amortized interest on the original $50,000 principal of the loan. For the duration of the loan, the consumer makes payments to the bank and no other party is involved in the transaction.

In contrast to a loan, a bond is a financial instrument which represents a promise to pay a fixed amount shown on the bond (“par value”) at a fixed future date to whoever happens to be holding that bond on that date. The entity (typically a corporation or government) borrowing the money may or may not make incremental interest payments to the bondholder during the life of the bond prior to its maturity date. Because the final payout at maturity is a fixed amount (the “par value”), the current price of the bond is always LESS than that par value. That may take some thinking to wrap your head around it. As an example, a simple bond without any intermediate “coupon payments” will be described.

If Company A issues a bond with par value $1,000 that matures five years from today, the financial terms of the bond are dictated SOLELY by those parameters – the par value, the corporation issuing the bond and the maturity date. The INTEREST RATE the company will pay is not explicitly identified on the bond itself, it is reflected in the price paid by the bondholder. If that $1,000 bond with 60 month maturity sells for $872 dollars today, the bondholder is essentially handing over $872 today in exchange for $1000 in 60 months and receiving $128 in interest on principal of $872. That’s 14.7% interest total but over 5 years, that’s the equivalent of only 2.775% interest compounded annually.

When the bond is issued, the corporation will collect the proceeds reflected by the price paid by the first person to buy the bond and that will reflect their interest rate on that “loan.” A firm with very good financial performance will likely net more from a bond sale than a firm with worse financial performance or bad credit. However, once the initial bond sales are closed, the firm has its money and faces no additional risk from fluctuations in that bond’s value. Obviously, if the borrower’s financial performance tanks and the value of its current bonds drop, future investors will be more wary about buying the entity’s NEXT round of bonds so borrowers cannot be oblivious to plight of current bondholders. (HOLD THAT THOUGHT.)

Bonds are a fungible instrument because rather than reflecting a single transaction with a single lender for the full amount of money being borrowed, they are fractionalized into standard amounts, typically $1,000, which allow a large “loan” of millions or billions of dollars to be spread out across thousands of individual investors, providing easier access to available funds.

Bonds can be very confusing because of the exponential math that drives their present value and the inverse relationship between price and interest rates. Small changes in interest rates can drastically alter market valuations of existing bonds. In the $1000 bond example above, if originally purchased for $872 in a world where interest rates were thought to be stable at 2.775%, then the $872 value five years out from maturity is a neutral deal. If interest rates jump to 4.0% the day after buying the bond for $872, the bond is now only worth $821. If the current bondholder who bought at $872 holds to maturity, they will still get their $1000 back and won’t lose that principal. However, if they purchased the bond hoping to sell it prior to maturity at a profit or simply provide a source of cash, the new $821 value represents a 5.8% loss in value from a 1.225% change in interest rates. If interest rates jump to 6%, that $872 bond would drop to $747, a 14.3% loss from a 3.22% change in interest rates.

Because of this magnification effect of bonds, bond traders routinely talk about changes to interest rates on bonds in terms of “basis points.” A basis point is simply one one-hundreth of a percentage point. An interest rate of 4.0 percent that jumps by five basis points is just 4.05% or 0.0405 as a decimal number. The use of this alternate interest rate unity with bonds is a subtle reflection that those trading in bonds for a living typically expect SMALL changes in interest rates over periods of days and weeks. One can subsequently infer that those trading bonds for a living are predisposed to freaking out when interest rates change by 20 or 30 basis points in a day. (HOLD THAT THOUGHT.)

While the math involved with calculating the value of a bond is complex and often un-intuitive, the simplicity of the bond’s core financial structure (corporate name, face value, maturity date) makes bonds simple to trade. The trading simplicity subsequently makes bond PRICES a highly transparent reflection of a variety of risks within financial markets. Specifically,

  • unique risks of default of the issuing company
  • risks unique to the issuing company’s industry
  • risks associated with the issuing company’s primary markets
  • inflation risks in the currency used to price the bond

The overall global bond market is estimated to be worth around $140 trillion dollars while the global stock market is estimated at roughly $115 trillion so it can be argued bonds reflect uncertainties in a much wider swath of financial activity than stocks and do so with a bigger magnifying glass.

Bonds also hold a position of particular influence on markets because one of the biggest issuers of bonds in the world economy is the American federal government, which has accumulated a total of $36 trillion dollars in debt. All of that debt is in the form of Treasury bonds so the American federal debt itself amounts to a staggering 25.7 percent of all bonds worldwide. Prior to the Trump Administration of 2025, no American President had ever called into question the “full faith and credit” of the United States in its obligation to pay back bonds. That pledge, coupled with the obvious economic power of the American economy, has led to Treasury bonds being used as the benchmark of financial safety for any other lending transaction, certainly in the United States but to some extent, the entire world. If markets currently have Treasury Bills selling at prices reflecting 4.0 percent interest rates, no other borrower is likely to be able to borrow money at rates below 4.0 percent because no other entity would be perceived as safer than the US government. (HOLD THAT THOUGHT.)


Fractional Reserve Banking

When a customer opens a savings account with $10,000 dollars, the bank doesn’t just put the money in the vault waiting for the customer to eventually withdraw the money. Having money sit idle produces no income to pay for the cost of operating the vault, much less generating income to the bank as compensation for providing the vault service. In order to produce income from customer A’s deposit, the bank actually attempts to loan that money to other customers. When another customer signs a loan paying 7% interest on a new car for six years, the bank takes that 7% in interest and uses part of it to cover costs of running the bank and part of it to depositors in order to attract their money so the bank can make the car loan.

But if the bank accepts $10,000 in cash from five depositors and makes a $50,000 loan to that car buyer, it winds up with no cash on hand. How will it be able to handle a sudden withdrawal of $2000 from one of the depositors? Banks are prevented from “over-lending” by enforcement of a minimum reserve ratio. The reserve ratio reflects the fraction of total deposits that must be kept physically on hand as cash. A higher ratio enforces a more conservative (“constructively paranoid”) amount of cash to be kept on hand but higher ratios limit the amount of lending the bank can originate. Lower ratios increase the total value of loans the bank can make – which consumers might like since loans would be easier to obtain at lower interest rates – but does so at higher risk if depositors ever generate an unexpected spike in withdrawals.

Even with a reserve ratio, the magnification effect of fractional reserve banking is often shocking to average citizens unfamiliar with banking mechanics. The process seems to be creating money out of thin air. In a very real sense, it is. In a town with a single bank operating with a reserve ratio of 0.2, an initial deposit by one customer of $10,000 can produce a total of $50,000 in “money” across all accounts. How?

  • initial deposit from A of $10,000
  • bank keeps 20% of that initial deposit on hand and loans the other $8,000 to B as a loan
  • B uses loan to pay $8000 for a car from C
  • C deposits the $8000 paid by B into an account at the bank
  • the bank now has $18,000 in deposits and can loan 80% of the additional $8000 in account C to another customer
  • repeat this process over and over

Repeating this process of deposits, loans and new deposits in this single closed system results in the following stream of dollars being added to the bank’s deposits:

$10,000... $8,000... $6400... $5120... $4096... $3276...

Because of the less than 1.0 fraction, this series of numbers will tend to zero after about twenty terms but the SUM of all of those numbers will be exactly equal to $10,000 / r where r is that reserve ratio. For a reserve ratio of r = 0.20, a $10,000 deposit results in $50,000 of total money in the banking system.

Average citizens in most modern economies are surprised when the mechanics of fractional reserve banking are thus explained with a 0.20 or 0.10 ratio as an example. The actual mechanics are even more shocking, even for those familiar with the basic concept. Historically, reserve ratios weren’t 0.20 or even 0.10. Reserve ratios followed by banks evolved to being around 0.05 for large banks. That means a $10,000 deposit results in a staggering $200,000 of total money in the system. But in reality, in the United States, the Federal Reserve eliminated reserve requirements on “time deposits” (savings accounts, CDs or other accounts with a fixed term expectation of how long money would remain in the account) in 1990 and eliminated them on ALL account types as of March 26, 2020 as part of efforts to stimulate the economy during the initial COVID meltdown. These zero reserve ratio requirements reflect a larger shift in Federal Reserve policy that assumes interest rates are a better mechanism for restraining or juicing the economy versus limiting the money multiplier effect of reserve ratios.

The key takeaway on fractional reserve banking and reserve ratios is that even though banks may not face an absolute requirement to meet or exceed some minimum reserve ratio, most banks don’t attempt to operate with zero reserves every day. However, the bar they must meet is only internally imposed and isn’t imposed by their regulator. That has another implication which will be raised as part of overnight lending.


Overnight Lending

The ability to create money out of thin air based upon numbers on balance sheets is a huge temptation to individual bankers and entire financial institutions. As part of ongoing regulation of this power, every bank in the system must settle its books every night and “prove” to its regulating body that liabilities match assets to the penny and that every dollar claimed to be present in every account is accounted for. In America, in olden days prior to March 26, 2020, banks were also required to confirm physical cash on hand satisfied applicable Federal Reserve reserve ratios. Since customers are depositing and withdrawing money nearly continuously, banks pick a point each evening after “lobby hours” (say 10:00pm) and take a snapshot of every account balance and use those balances as the starting point in this settlement process. Inevitably, the numbers don’t exactly match so a bank short on on-hand cash is required to borrow the difference from other banks.

A local bank with five thousand customers with average balances of $5,000 would thus have $25,000,000 in deposits and might find a nightly “close” would turn up a shortfall of $150,000 because ten customers took advantage of a sale at the local car dealer and each withdrew $15,000 to use in buying a new car. That unexpected drop in cash might leave the bank mismatched between assets and liabilities or below its target internal reserve ratio target, requiring it to borrow money from another bank for the night until the next day’s transactions help settle the imbalance. The shortfall is only 0.6 percent of total deposits so it isn’t a crisis yet the bank is obligated to find $150,000 immediately in order to balance its books before opening the next day.

Like anyone else borrowing money for any period of time, the bank wants to find a lender willing to lend the needed amount at the lowest rate possible. But like any other loan, the lender wants to ensure the borrower is good for the money and wants to charge the borrower an interest rate commensurate with the risk posed by that borrower. Because these loans typically have overnight or seven day terms and because the borrower IS a bank, these overnight loans usually involve little drama and relatively low interest rates.

The availability of these extremely short term loans and the relatively low interest rates thus act as “motor oil” in the larger financial engine, keeping the entire system running smoothly. However, no bank has infinite resources to lend to anyone, including fellow banks, and even banks look at other banks with suspicion when LARGE amounts of money are needed urgently. This makes the interest rates charged for overnight loans a key indicator of the underlying health of the entire financial system.


Overnight Rates

As of 2025, there are two key interest rates that reflect the stresses being worked out in overnight loan each day. One rate is calculated by the Federal Reserve and termed the EFFR (Effective Federal Funds Rate) and reflects the interest charged by the Federal Reserve to member banks for loans those member banks obtained WITHOUT collateral. The other rate is SOFR (Secured Overnight Financing Rate) which is published by the New York Federal Reserve Bank after collecting data on rates charged on loans in the repurchase (“repo”) markets that help banks settle imbalances by trading Treasury Bills. The distinction between these two rates is crucial to understanding the psychology in this nightly lending activity.

The SOFR rate is derived from transactions that involve Treasury Bills as collateral, meaning those “repo” loans are normally thought to be the absolute safest, low-risk loan that can possibly be made. The borrower is a BANK. The term of the loan is exceptionally SHORT (a day or a week), making the time risk exceptionally small. And the borrower is offering up what should be the safest financial instrument, T-BILLS, as collateral.

In contrast, the EFFR reflects loans made by the Federal Reserve to banks who provided no collateral for the loan whatsoever. It’s still a bank borrowing the money, the term of the loan is still exceptionally short but the borrower is offering no collateral. The risk of these loans is logically higher than collateralized loans and the interest rate charged will be higher.

Circling back to the points made in the section on bond behavior, the process of setting overnight interest rates for inter-bank lending is directly influenced by current bond rates, most notably interest rates on US Treasury bonds. Why? Again, axiomatically, in financial markets across the globe, the “full faith and credit” pledge of the American government regarding its bonds makes Treasury bills the ultimate benchmark of safety. If rates on Treasury bills go up, interest rates charged on overnight loans will IMMEDIATELY rise as well, to levels that will always be higher than those benchmark Treasuries. That means fluctuations in the perceived risk of US Treasuries will add friction to this nightly settlement process, which can be further magnified by unexpected shocks in the market.

At this point, it is also worth tying together two other points about overnight lending and interest rates reflecting risk. Since US banks have ZERO externally imposed reserve requirements actually being enforced by the Federal Reserve, any borrowing they undertake as part of their nightly settlments is being intiated for one of two reasons. The bank either needs the money simply to match up assets and liabilities (that requirement doesn't going away with zero reserve requirements) or they need it to satisfy internal corporate goals for their reserve ratio to avoid an even tighter squeeze in the future. If banks are seeing overnight lending come close to locking up with fewer banks willing to lend, either there are less reserves in the entire system to borrow from and/or those with reserves to lend are increasingly concerned about the health of those needing to borrow and are driving up interest rates accordingly, making the borrowing more expensive at the time it is needed most urgently. This is not a good sign when banks are already operating with virtually no limits imposed by the Federal Reserve.


The Crash That Wasn’t

All of the prior sections provided background on a complex financial system reflecting:

  1. a banking system whose participants are required to settle their books EVERY DAY by their regulator
  2. bond markets optimized to identify borrowing risk and instantly / continuously reflect it in bond prices
  3. a banking system whose regulator in the United States has virtually removed all external pressure on regulated banks to meet even the lowest of reserve ratios when deciding how much cash to keep on hand
  4. nightly processes presumably executed by professionals who are the most comfortable with handling high pressure transactions with large sums of money with ever-fluctuating levels of risk across a variety of factors
  5. regulatory processes that generate publicly shared statistics that summarize the risks seen by the professionals handling the deals required by these nightly reconciliations

With all of that background, the events of April 9, 2025 when the Trump Administration “paused” all previously announced tariffs except for those applied to China can be analyzed with better perspective.

As of the early evening of April 8, 2025, no one in the Trump Administration was publicly admitting to consideration being given to pause, alter or eliminate any prior announced tariffs on any country. In fact, on April 6, Administration officials explicitly denied a “news report” stemming from a tweet on X from a reporter no one had ever heard of that claimed Trump was considering a 90-day pause on all tariffs except China while reviewing the overall strategy. The market had surged on the initial news then dropped $2.4 trillion in eight minutes after Trump officials dismissed the story.

On April 7, Administration officials had appeared on multiple outlets sending decidedly mixed messages, some saying Trump would decide what Trump wanted to do, others stating no “negotiations” were underway with any countries. Trump himself appeared at a Republic Party event patting himself on the back for supposedly fielding multiple calls from various countries “kissing his ass” to roll back tariffs on their country.

As of April 10, it appears there WAS no plan for pauses or rollbacks as of Trump’s appearance at the Republican event. However, afterwards, administration officials were getting input from people familiar with bond markets that things were awry. A Yahoo story on April 9, 2025 published after the one-day rebound quoted a user identified as “market veteran Jim Bianco” from a post on X: “We are seeing a disorderly liquidation.” Liquidation is probably not at the top of the list of words anyone involved with global markets wants to hear.

The theory presented in that story boils down to this:

  1. bond prices often RISE when stock prices decline by large amounts
  2. this inverse relationship reflects a “flight to safety” from equities that are falling in value to bonds whose prices RISE as more investors demand them to transfer money out of losing stocks
  3. yet bond prices were FALLING and yields on bonds were RISING as the market slumped on Tuesday
  4. bond investors were clearly expressing some larger concern about what was previously assumed to be a predictable relationship between stocks and bonds and risk as reflected by interest rates

The Yahoo story further references sources that point to the likelihood that institutional investors (hedge funds, typically) had been trying to extract short term profits by setting up certain trades attempting to leverage small differences in short term and mid-term interest rates and rates fluctuated in a way not anticipated by those traders, leaving them short on certain positions requiring them to SELL bonds to raise cash to cover other positions. This selling volume was LOWERING prices on those bonds and RAISING interest rates, exactly counter to expectations which triggered more selling.

Remember the points made earlier about bonds being a "magnifier" of underlying variability because of the exponential math of compound interest over years. Now remember the point about reserve ratio requirements being eliminated by the Federal Reserve in 2020. Now remember the point about stocks dropping $2.8 trillion dollars in eight minutes on Monday after seeing a rumor of a pause in tariffs being denied by Trump officials.

Going into Tuesday night, the Trump Administration had not only re-iterated its commitment to sticking with the announced tariff regime, it announced an increase in the tariff on Chinese imports to 104 percent. And that was AFTER the Dow Jones Industrial Average dropped 3.0% during the day Tuesday. In total, the Dow had dropped from 41,736 on April 2, 2025 to 37,663 on April 8 solely due to Trump decisions on tariffs.

The net result was that bond traders were already seeing signs Tuesday night that highly leveraged bets on interest rate spreads were being destroyed due to the unprecedented shocks injected into the entire world economy by Trump’s arbitrary tariff decisions and additional shocks could trigger even larger leveraged positions to come unraveled, leading to another financial crisis. How big might that crisis be? Data on derivatives is NOTORIOUSLY hard to find and virtually impossible to verify (hence, a large reason why derivatives are still such a concern) but consider these statistics:

  • an article published in November 2008 estimated the total value of derivatives as of June 30, 2008 to be $683 trillion dollars (see https://www.bis.org/publ/otc_hy0811.pdf )
  • an estimate from June 2024 from the same organization estimated the value of derivatives to be $730 trillion
  • other estimates can be found placing the 2025 estimate at $1 quadrillion dollars

The Crash That May Still Come

The point of this entire post is that the American economy didn’t permanently dodge the big one when Trump announced a “pause” in the application of tariffs. The most plausible proximate cause for this near miss was concerns over bond trading from unexpected changes in interest rates upending highly leveraged hedge fund bets. But there are a multitude of growing issues that could all spike interest rates up or down in similar ways that could trigger equally large surprises for traders.

  1. Trump has not even technically eliminated the insane tariffs, he’s only “paused” them for 90 days – even if the tariffs are reduced to a “mere” 10% for most countries, that is still a huge inflationary shock to the US economy whose stock market was previously priced for perfection and lower inflation.
  2. Trump remains in power and likely to shoot off his mouth at random about any particular economic decision.
  3. Insiders within the Trump Administration have demonstrated a universal lack of ethics and willingness to violate their oath to uphold the law.
  4. The Trump DOJ has explicitly disbanded the National Cryptocurrency Enforcement Team and ordered the Market Integrity and Major Frauds Unit within the DOJ to stop investigations related to cryptocurrencies.
  5. The Trump Administration established a policy calling for a permanent reserve of Bitcoin – no purchases have been made yet, however the US already owns 198,109 BTC as of 1/15/2025 worth $19.21 billion which were seized as part of criminal investigations.
  6. It is no longer a given in world markets that the United States is committed to paying its debts with the full faith and credit of the United States.
  7. One of the largest external holders of US debt is China, which currently owns $761 billion worth of Treasury bills.

When the combination of a completely incompetent, corrupt and petulant President is added to an environment that has demonstrated to the world there is no adult with an IQ above 65 in remote proximity to the steering wheel on the US economy, investors worldwide have legitimate reason to question the full faith and credit of the United States, altering decisions in nearly every lending decision about what a “safe” interest rate should be. Investors know China could retaliate against the US by dumping large numbers of its Treasury holdings, tanking their price and consequently spiking US interest rates which would add tens of billions to our yearly deficit in the form of higher interest payments on the larger $36 trillion in total debt. Investors have legitimate reasons to believe the integrity of US markets will be compromised by the complete elimination of resources devoted to protecting the integrity of financial reporting relied upon by investors. Investors have legitimate reasons to think markets are being manipulated on an hourly basis by administration insiders who think they know where Trump is headed and can spook the market in the opposite direction to capture both sides of a giant swing. And investors know administration insiders likely believe there is zero chance of them being charged and prosecuted for such crimes and would be pardoned anyway as long as they swear fealty to Trump. Investors likely know INVESTOR insiders who are freaked out about other looming highly leveraged positions that could go wrong with another large shock applied to financial markets. And investors can see that the current US government has zero intent to regulate cryptocurrencies, virtually assuring future FTX-scale frauds will occur and cascade into meltdowns in traditional markets.

In light of the actions taken since January 20, 2025 and in light of the volatility seen in markets thus far, the question regarding “the big one” doesn’t seem to be an “if” style question. Only a “when.”


WTH

Meaning-Minimizing Memes

Content masquerading as news or commentary has been degrading actual informational quality since Fox News industrialized the broadcasting of propaganda that looked and tasted like “news” nearly 30 years ago. However, the advent of social media formats hyper-optimized for absurdly short texts, tweets or video shorts has made the problem exponentially worse. Short-form focused media requires ANY language included with it to shoulder a much higher burden in attracting attention (any attention, likes or dislikes). The result is language that attempts to put already ineffective “meme” type language on steroids, resulting in references like these:

Representative A Claps Back at Representative B
Senator C Shuts Down Senator D
Person E Called Out by Person F
Actor G EXPOSES Shocking Fraud
Senator H Owns J
Billionare K Destroys L
Epic Takedown of M by N
Celebrity P Dragged on Twitter after Comments

Claps back? Shuts down? Called out? EXPOSES? Owns? Destroys? Epic? Dragged?

Duuuuuuuuude….

Are the “creators” pushing this content twelve years old? This is the ultimate in bad language because it catastrophically fails at two levels. First, on the surface, this language SELDOM if ever accurately portrays the nature of the interaction being covered. It is usually overstating the impact of the exchange by an order of magnitude in terms of what was accomplished during that exchange or what was prevented from happening in that larger context. Those involved are often politicians. It's thus already clear many have no shame in being caught lying or pushing corrupt agendas of their PAC donors.

More importantly, though, these passive aggressive terms used so commonly in personal communications reflect problems with the social mindset of people addicted to social media. We have over a decade of experience with “Karens” (not exclusively female...) loose in the wild. Karens are people who have been fed a highly curated stream of negative content designed for and targeted towards them to make them feel superior to others while supporting extreme attitudes and policies driven by ignorance and hate. A continuous diet of such content not only makes Karens stupid and hateful, it acclimates them to screaming back through their screen at all those “others” as they digest hours of that content every day. Of course, that feeling of entitlement to hate others spills into real life on the street when they encounter actual “others” they’ve been hating at home for months.

But Karen behavior isn’t the only problem created by 24x7 diets of curated social media content. Imagine a day where news of a committee meeting on Capital Hill pointed out a looming disaster in the operation of a government agency or a gross violation of a citizen’s rights or a blatantly corrupt elected official. Before social media technologies existed, a citizen hearing that news might have been tempted to

  • write a paper letter to their elected official
  • write a paper letter to the elected official involved with the hearing (even if not their elected official)
  • share the story with like-minded friends via email, letter or phone call
  • sit back and do nothing, saying “I hate politics”

In the current era polluted by AI generated content and automated fake likes, citizens still have all of the old tools above for responding to such news at their disposal, but they also have new ones unique to social media:

  • like or upvote the snippet that drew their attention
  • retweet / forward that snippet under their online identity / avatar
  • do nothing, assuming that the idiotic official who needs a head slap just got it via the negative social media attention
  • send an email to their elected official via a restrictive web form on a Congressional web site which likely gets electronically sorted and counted and never read directly
  • send an email to the elected official involved with the hearing (even if not their own) via a restrictive web form on a Congressional web site which likely gets electronically sorted and counted and never read directly

For millions who grew up with social media platforms as children and who likely view online references to them as the worst possible “dis”, it seems likely that many citizens assume a pithy, pugnacious tweet is enough to “shame” a bad guy, help encourage a downdraft in that bad guy’s public personna and eventually stop whatever they were trying to do. Internet karma will eventually achieve their desired result. It’s tempting to think that, when you stare at your phone for four hours a day and see fifty tweets or video links with these titles, surely the public has caught on. But remember, much of the content is auto-generated and a large share of the likes and comments related to those content items are auto-generated as well. There's no guarantee that ANY particular item seen online was generated by a human who was "engaged."

Unfortunately, as already mentioned, people attracted to money and power often lack any of the shame chromosomes present in normal humans and are virtually immune to feelings of regret after being “called out” or “dragged” on social media platforms. More importantly, Congress and the courts don’t operate on a “like” basis nor care who gets “owned” on X or BlueSky or Instagram. Those getting dissed on social media MIGHT eventually fall out of favor and lose power but they can still do an enormous amount of harm until internet karma catches up.

The reality is that all of this “owning” and “clapping back” almost NEVER succeeds at meaningfully stopping an opponent in the short term yet that citizen has likely been pacified to the point of not doing anything. This pacification is not the goal of those creating content with these childish language memes but it is VERY likely the exact outcome they are producing with their work. And this apathy and inaction is EXACTLY what those on the other side of that issue want to see.

If you’re a member of the chattering class, the commentariat, a podcaster regurgitating other people’s work as new insight, etc. and you wrap your “content” with this type of terminology, STOP. Your “content” might be collecting views and likes and subscribers but it is NOT likely triggering true engagement where it matters. You are not educating your content consumer, you are simply lulling them into thinking justice has been served and no further action on their part is required.

If you frequently read or watch content that uses this type of language incessantly, you need to find better sources of news and opinion. Any source relying on this type of language is reinforcing (for one side or the other) the existing tribalism that allowed the masses to be more easily separated from their own money. Sites using this type of language are highly unlikely to be actually explaining the mechanics of an issue and how possibly both “sides” are failing to frame an appropriate solution that doesn’t entrench the status quo.


WTH