Last year when you wrote about the SVB debacle you mentioned the mark to market method of valuation. When I ask Perplexity about the purpose of the Fed it answers, in part: "The primary purpose of the Federal Reserve System is to provide the United States with a safer, more flexible, and more stable monetary and financial system." At least that's the *stated* purpose. I suppose the gov't and bank bedfellows relationship could almost be construed as a conflict of interest. Always interesting to learn more of our byzantine financial system. (BTW, the link for Andreessen & Horowitz isn't working).
In the early days of the web, I'd say most developers were mostly focused on getting things to work asap, getting on the market to achieve first mover market share advantage and/or beating the competition, growing users fast, to display and function as intended, and with as much uptime and as few crashes and as little lag as possible. Security in its various respects took second place and was not given priority emphasis by most developers. That includes the security of protecting proprietary intellectual property (about which there was a commonly-held aversion in the small-scale programmer culture back then), but also in terms of information assurance, robust identify verification, access controls, and so forth.
"View Source" was thus a valuable exercise at that time, for developers and hackers alike. Even when things like java appeared in those wild days it was often possible to just download the source code for those programs and libraries, and best practices on the web had not yet gotten so very complicated or relying on many layers of abstraction and intermediary tools such that a developer of ordinary talent could actually read and understand a lot of that early code and thus learn and apply the lessons and techniques in their own efforts.
Before "maximum monetization", many websites - for example for newspapers - were extremely lean and parsimonious with storage and bandwidth and young people will simply not believe you today that, for a display of content that appears the same to the end user, many of those websites loaded faster a quarter of a century ago because today they are completely larded up with thousands of times greater size, complexity, network and database calls, auxiliary functions, incredibly obnoxious surveillance, etc. I hate to say it, but Substack's approach - while I'm sure they can defend themselves with plenty of good reasons why it has to be this way in this day and age - is an example of just how bad and crazy things have gotten in terms of what it takes to produce what is in effect an incredible bloated resources hog and elaborate virtual simulation of the once-simple and ancient concept of a basic internet discussion forum / blog.
Those old conditions are all long gone today, and, as I understand it, for many years this has gone all the way down to secret proprietary security protections on the CPUs themselves and on other important chips. While developers may still share some of their AI / LLM tips and tricks with each other for a while, the mindset of the whole industry has matured and evolved and the stuff that requires a lot of time and money to develop will still likely be kept under wraps to the extent the companies fronting the resources can do so.
One of the problems with AI is that it isn't 100% replicable. Following steps A, B and C gets you result D1 D2 D3 when 3 different people try it. They are probably similar but they may not be and the differences could be critical
I hope Marc Andreessen and Ben Horowitz are wrong but I'm inclined to believe them. That said, it seems worth noting they provide no evidence to back up their claims, not even anecdotes. Their piece reminds me a lot of a charity talking about how bad the environment, food, shelter, illness, etc. at the center of their concern is. I also wonder if they even have framed the issue accurately. For example, are regulators actively dissuading banks from issuing loans to startups? Could it be they are instead setting standards or guidance that most startups happen not to meet and it isn't intended to harm startups as MA and BH make it sound? What I'd really like to see is regulator statements showing intent or statistical data that verifies changes beyond just anecdotal cases.
Ignoring concerns that scotus gave itself more power, I don't know why more judicial oversight of executive implementation of ambiguous laws makes anything faker or less enforceable.
The Biden administration purported to be able to regulate AI by executive action without statutory authorization. Under Loper, it's much tougher to pull that off than under Chevron, because under Chevron you could gesture to "gaps" in the statute that the court could determine to be ambiguities, but that practice is now in much greater disfavor. So I can understand why they are complaining about M&A scrutiny, but gesturing to overall regulatory issues impacting software startups seems more like an excuse.
I still don't really see your first comment clearly but I get your point now and mostly agree.
The only contrary point to note is that the executive branch can continue to do exactly the same as before. It's just that they are somewhat more likely to lose if and when their regulation is challenged in court. Before it had to go to court, be contrary to law, the court rule that way. Now it has to go to court, be in the gray, AND the court has to disagree with their regulation.
No comments on Ed West's post after a day? Sad (as the orange man would say).
The Brits have welcomed immigrants from far-away places in such numbers that they are now in position to expect British law to conform with values that have no basis in British history or traditions. Open borders + democracy = it's not your country anymore. Should Americans take note and act accordingly? Nah - If we do we get called nasty names and who can stand up to that?
These days most apps are written in React or another client-side framework, so view source isn't as useful. And of course mobile apps don't have that at all.
For me a very serious mistake is forgetting that IA renaissance began with reinforcement learning, not LLM. I still think that reinforcement learning will be more important. It will the basis of swarms, and I has already given us Alpha Fold.
As an economist, it is the technology that breaks the curse of dimensionality:
I do not have anything like a rigorous proof of this proposition, but my own hunch and impression is that at the highest level of abstraction and at the extreme limits of scale and fractal granularity, that there is some kind of Fundamental Equivalence Theorem which would show that all effective ways of extracting signals from noise, of recognizing patterns, of making accurate predictions or generating typed content, doing statistical regression, etc. are, in the final analysis, all the same thing, indeed, perhaps at sufficient scale and level of detail, even the "One Neat Trick" behind what we call "intelligence" itself. What we imagine are many ways to skin the cat are really at root all the same abstract approach pursued perhaps with different levels of efficiency and error tolerance, and so with enough "compute" all roads lead to Rome.
I get that capital improves a bank's ability to weather adversity but it seems like the cost of deposits (interest rate and operating costs) versus the interest and gains on loans and other investments would be at least as important. Maybe it would also be worth considering how much on each side of the ledger is fixed rate and how much adjusts. Of course that gets more complicated with mortgage loans that get refinanced when rates rise. Anyway, do regulators look at more than just the value of holdings and capital ratio? I don't think I've ever heard more mentioned.
Are you sure that capital requirements improve a bank's ability to weather adversity? It seems an argument could be made such requirements might actually worsen that ability.
Requiring banks to hold capital — rather than lend it at attractive rates — would be expected to bring down banks' average return on capital. This would incentivize management to reach for increased yield on the funds available for lending to bring that average back up.
It seems if we want to improve fiscal responsibility, the focus should be on loan underwriting quality. To provide real incentives for this, legislation could be enacted making at least a part of managements' personal assets available to creditors where irresponsible underwriting occurred. This would have the added benefit of freeing up capital for economic growth.
First, with the exception of rare cases such as the opioid settlements, any company's management's personal assets are not at risk to creditors. As much as anything, that is why we have companies.
Second, to the best of my knowledge, what you are referring to is the reserve requirement. Capital requirements are primarily equity requirements and secondarily most debt other than federally insured deposits, both of which would be at risk if the bank fails. This is similar but different to what you are asking for.
It is really odd that banks should be holding a lot of long term fixed-rate assets that would NEED to be marked to market. Managing transformation risk is OK, but interest rate risk???
My own opinion is that it's not odd. I think the banks were doing what they were told to do, knew it was risky bad practice, but knew that if it went bad that their taskmaster patron (Uncle Sam) would still bail them out.
The big banks are not exactly 100% mere extensions of the government and obedient channels for the implementation of state action and policy, but they are now much closer to that than genuinely "private" entities, and also symbiotic and co-dependent, or, to use Arnold's excellent term, "Government-Adjacent".
I don't think it's exactly right - or at least not complete - to describe the current system as merely "privatized gains, socialized risks." It's not just "Too Big To Fail" moral hazard.
Instead, the government wants people to have access to 30-year fixed rate mortgages*, and it wants to fund its deficits by selling all those 10 and 30 year fixed rate bonds, and it NEEDS the financial sector to go along with this stuff without balking, even though everyone understands rate risk and other risks and the real possibility all of this explodes like a ticking time bomb.
And, it seems to me, the financial sector and the big banks DO go along with it, at rates that are lower and in quantities that are higher than they otherwise would. This is apparently because everyone seems to believe in some tacit understanding that if a bank goes bust in this way by doing the state's bidding - what the government wants, encourages, or insists that it do - that they were being "paid" or "compensated" for going beyond prudent fiscal limits in this regard by the state's "promise" to provide wink-wink "insurance" and to have their back and act as if to some extent it had extended some kind of a surety in the first place.
And that this is especially true if the reason for the shift in conditions that caused the ticking time bomb to go off was also arguably the government's "fault" and so it justly "ought" to be responsible for and help out the banks harmed by those decisions. With regard to rate risk, this would be like arguing that the recent spike in inflation resulted from government policy choices for over-stimulation, money printing, and soft monetary policy.
If I make it hard for you to do business if you don't buy and hold a lot of my bonds, then I debase the currency more than people expected me too, then your losses are indeed kind of my fault. Instead of publicly taking responsibility, I get to pretend this is all private greedy bankers making bad risky decisions and here is the innocent angel government forced to come in to clean up their mess because it has no choice if it is to prevent this little disaster from kicking off a big crisis where a lot of innocent people the government is charged with protecting are going to get hurt too. This seems to work out for the politicians, and so it's guaranteed to continue and probably even to keep expanding.
*As an interesting bit of trivia, total mortgage outstanding is now over $20 Trillion, and is only 37% higher than the peak of the "bubble" 16 years ago, a nominal annual growth rate of only 2%, and arguably still negative in real terms.
"Historical cost valuation is problematic because a large share of bank securities, loans and leases were acquired prior to March 2022, when interest rates were at historic lows. At today’s higher interest rates, these assets have lost significant market value, but since these losses have not been realized, they are not reflected in bank asset valuations or regulatory capital measures."
I don't read many econ or finance blogs at all - being over my head - but I feel like I read some pushback against this contention [e.g. B of A holding company: 0 capital] somewhere in the past year ... Is this situation regarding securities, loans really as straightforwardly dire as presented?
I think it is worth also noting that the Dickey amendment effectively outlawed federal funding for firearm-related research from 1997 to 2013, with small funding until 2020, more after, and lifting of restrictions just this year.
Unfortunately, I'm not hopeful lifting of the ban will result in any funding of research on gun benefits. It would be great if there could be a relatively unbiased look at injury prevention that considers defensive use, accidents and violent offenders (crimes of passion and habitual criminals) while minimally interfering with recreational use.
My browser (or the website, or something else ?) won't register my "Like" for this comment. But rest assured, Guest, I do like it. I'm convinced that legal gun ownership poses a significant deterrent to crime.
Last year when you wrote about the SVB debacle you mentioned the mark to market method of valuation. When I ask Perplexity about the purpose of the Fed it answers, in part: "The primary purpose of the Federal Reserve System is to provide the United States with a safer, more flexible, and more stable monetary and financial system." At least that's the *stated* purpose. I suppose the gov't and bank bedfellows relationship could almost be construed as a conflict of interest. Always interesting to learn more of our byzantine financial system. (BTW, the link for Andreessen & Horowitz isn't working).
fixed. thanks
The “view source” mechanism of AI appears to be people sharing lessons learned on Substack, Twitter, etc.
EDIT: See also Anthropic's 'artifacts' tool for Claude, about which there is some information here: https://news.lore.com/p/claude-makes-ai-easy-to-share-and-teach?utm_source=news.lore.com&utm_medium=newsletter&utm_campaign=claude-makes-ai-easy-to-share-and-teach
In the early days of the web, I'd say most developers were mostly focused on getting things to work asap, getting on the market to achieve first mover market share advantage and/or beating the competition, growing users fast, to display and function as intended, and with as much uptime and as few crashes and as little lag as possible. Security in its various respects took second place and was not given priority emphasis by most developers. That includes the security of protecting proprietary intellectual property (about which there was a commonly-held aversion in the small-scale programmer culture back then), but also in terms of information assurance, robust identify verification, access controls, and so forth.
"View Source" was thus a valuable exercise at that time, for developers and hackers alike. Even when things like java appeared in those wild days it was often possible to just download the source code for those programs and libraries, and best practices on the web had not yet gotten so very complicated or relying on many layers of abstraction and intermediary tools such that a developer of ordinary talent could actually read and understand a lot of that early code and thus learn and apply the lessons and techniques in their own efforts.
Before "maximum monetization", many websites - for example for newspapers - were extremely lean and parsimonious with storage and bandwidth and young people will simply not believe you today that, for a display of content that appears the same to the end user, many of those websites loaded faster a quarter of a century ago because today they are completely larded up with thousands of times greater size, complexity, network and database calls, auxiliary functions, incredibly obnoxious surveillance, etc. I hate to say it, but Substack's approach - while I'm sure they can defend themselves with plenty of good reasons why it has to be this way in this day and age - is an example of just how bad and crazy things have gotten in terms of what it takes to produce what is in effect an incredible bloated resources hog and elaborate virtual simulation of the once-simple and ancient concept of a basic internet discussion forum / blog.
Those old conditions are all long gone today, and, as I understand it, for many years this has gone all the way down to secret proprietary security protections on the CPUs themselves and on other important chips. While developers may still share some of their AI / LLM tips and tricks with each other for a while, the mindset of the whole industry has matured and evolved and the stuff that requires a lot of time and money to develop will still likely be kept under wraps to the extent the companies fronting the resources can do so.
Yup, software is much less open today than it was a few decades ago.
One of the problems with AI is that it isn't 100% replicable. Following steps A, B and C gets you result D1 D2 D3 when 3 different people try it. They are probably similar but they may not be and the differences could be critical
I wouldn't phrase that as a problem. It's just how LLMs work.
It's a feature not a bug. I know.
It is, IMHO, another reason why LLM AI is likely to be of limited use because there are many occasions where you want some kind of predictability.
It's not as bad as the hallucination problem which is actively dangerous (see https://ombreolivier.substack.com/p/llm-considered-harmful ) but it's another strike against it.
I hope Marc Andreessen and Ben Horowitz are wrong but I'm inclined to believe them. That said, it seems worth noting they provide no evidence to back up their claims, not even anecdotes. Their piece reminds me a lot of a charity talking about how bad the environment, food, shelter, illness, etc. at the center of their concern is. I also wonder if they even have framed the issue accurately. For example, are regulators actively dissuading banks from issuing loans to startups? Could it be they are instead setting standards or guidance that most startups happen not to meet and it isn't intended to harm startups as MA and BH make it sound? What I'd really like to see is regulator statements showing intent or statistical data that verifies changes beyond just anecdotal cases.
The Biden AI regulation was pretty fake and unenforceable even under Chevron, and now it's even faker and less enforceable under Loper Bright.
Ignoring concerns that scotus gave itself more power, I don't know why more judicial oversight of executive implementation of ambiguous laws makes anything faker or less enforceable.
The Biden administration purported to be able to regulate AI by executive action without statutory authorization. Under Loper, it's much tougher to pull that off than under Chevron, because under Chevron you could gesture to "gaps" in the statute that the court could determine to be ambiguities, but that practice is now in much greater disfavor. So I can understand why they are complaining about M&A scrutiny, but gesturing to overall regulatory issues impacting software startups seems more like an excuse.
I still don't really see your first comment clearly but I get your point now and mostly agree.
The only contrary point to note is that the executive branch can continue to do exactly the same as before. It's just that they are somewhat more likely to lose if and when their regulation is challenged in court. Before it had to go to court, be contrary to law, the court rule that way. Now it has to go to court, be in the gray, AND the court has to disagree with their regulation.
No comments on Ed West's post after a day? Sad (as the orange man would say).
The Brits have welcomed immigrants from far-away places in such numbers that they are now in position to expect British law to conform with values that have no basis in British history or traditions. Open borders + democracy = it's not your country anymore. Should Americans take note and act accordingly? Nah - If we do we get called nasty names and who can stand up to that?
Ken
These days most apps are written in React or another client-side framework, so view source isn't as useful. And of course mobile apps don't have that at all.
For me a very serious mistake is forgetting that IA renaissance began with reinforcement learning, not LLM. I still think that reinforcement learning will be more important. It will the basis of swarms, and I has already given us Alpha Fold.
As an economist, it is the technology that breaks the curse of dimensionality:
https://forum.effectivealtruism.org/posts/uHeeE5d96TKowTzjA/world-and-mind-in-artificial-intelligence-arguments-against
I do not have anything like a rigorous proof of this proposition, but my own hunch and impression is that at the highest level of abstraction and at the extreme limits of scale and fractal granularity, that there is some kind of Fundamental Equivalence Theorem which would show that all effective ways of extracting signals from noise, of recognizing patterns, of making accurate predictions or generating typed content, doing statistical regression, etc. are, in the final analysis, all the same thing, indeed, perhaps at sufficient scale and level of detail, even the "One Neat Trick" behind what we call "intelligence" itself. What we imagine are many ways to skin the cat are really at root all the same abstract approach pursued perhaps with different levels of efficiency and error tolerance, and so with enough "compute" all roads lead to Rome.
I get that capital improves a bank's ability to weather adversity but it seems like the cost of deposits (interest rate and operating costs) versus the interest and gains on loans and other investments would be at least as important. Maybe it would also be worth considering how much on each side of the ledger is fixed rate and how much adjusts. Of course that gets more complicated with mortgage loans that get refinanced when rates rise. Anyway, do regulators look at more than just the value of holdings and capital ratio? I don't think I've ever heard more mentioned.
Are you sure that capital requirements improve a bank's ability to weather adversity? It seems an argument could be made such requirements might actually worsen that ability.
Requiring banks to hold capital — rather than lend it at attractive rates — would be expected to bring down banks' average return on capital. This would incentivize management to reach for increased yield on the funds available for lending to bring that average back up.
It seems if we want to improve fiscal responsibility, the focus should be on loan underwriting quality. To provide real incentives for this, legislation could be enacted making at least a part of managements' personal assets available to creditors where irresponsible underwriting occurred. This would have the added benefit of freeing up capital for economic growth.
First, with the exception of rare cases such as the opioid settlements, any company's management's personal assets are not at risk to creditors. As much as anything, that is why we have companies.
Second, to the best of my knowledge, what you are referring to is the reserve requirement. Capital requirements are primarily equity requirements and secondarily most debt other than federally insured deposits, both of which would be at risk if the bank fails. This is similar but different to what you are asking for.
I understand. That is why I suggested legislation would be required to make that change for bank managements.
Not sure what you say you understand
That management's personal assets are usually not at risk to creditors.
It is really odd that banks should be holding a lot of long term fixed-rate assets that would NEED to be marked to market. Managing transformation risk is OK, but interest rate risk???
My own opinion is that it's not odd. I think the banks were doing what they were told to do, knew it was risky bad practice, but knew that if it went bad that their taskmaster patron (Uncle Sam) would still bail them out.
The big banks are not exactly 100% mere extensions of the government and obedient channels for the implementation of state action and policy, but they are now much closer to that than genuinely "private" entities, and also symbiotic and co-dependent, or, to use Arnold's excellent term, "Government-Adjacent".
I don't think it's exactly right - or at least not complete - to describe the current system as merely "privatized gains, socialized risks." It's not just "Too Big To Fail" moral hazard.
Instead, the government wants people to have access to 30-year fixed rate mortgages*, and it wants to fund its deficits by selling all those 10 and 30 year fixed rate bonds, and it NEEDS the financial sector to go along with this stuff without balking, even though everyone understands rate risk and other risks and the real possibility all of this explodes like a ticking time bomb.
And, it seems to me, the financial sector and the big banks DO go along with it, at rates that are lower and in quantities that are higher than they otherwise would. This is apparently because everyone seems to believe in some tacit understanding that if a bank goes bust in this way by doing the state's bidding - what the government wants, encourages, or insists that it do - that they were being "paid" or "compensated" for going beyond prudent fiscal limits in this regard by the state's "promise" to provide wink-wink "insurance" and to have their back and act as if to some extent it had extended some kind of a surety in the first place.
And that this is especially true if the reason for the shift in conditions that caused the ticking time bomb to go off was also arguably the government's "fault" and so it justly "ought" to be responsible for and help out the banks harmed by those decisions. With regard to rate risk, this would be like arguing that the recent spike in inflation resulted from government policy choices for over-stimulation, money printing, and soft monetary policy.
If I make it hard for you to do business if you don't buy and hold a lot of my bonds, then I debase the currency more than people expected me too, then your losses are indeed kind of my fault. Instead of publicly taking responsibility, I get to pretend this is all private greedy bankers making bad risky decisions and here is the innocent angel government forced to come in to clean up their mess because it has no choice if it is to prevent this little disaster from kicking off a big crisis where a lot of innocent people the government is charged with protecting are going to get hurt too. This seems to work out for the politicians, and so it's guaranteed to continue and probably even to keep expanding.
*As an interesting bit of trivia, total mortgage outstanding is now over $20 Trillion, and is only 37% higher than the peak of the "bubble" 16 years ago, a nominal annual growth rate of only 2%, and arguably still negative in real terms.
My own guess is that it is more a sin of omission -- regulators are too young to remember the S&L crisis -- than commission.
I would point out that if the government wants to fund its own LT bonds more cheaply, it is a mistake to encourage 30-year fixed-rate mortgages.
"Historical cost valuation is problematic because a large share of bank securities, loans and leases were acquired prior to March 2022, when interest rates were at historic lows. At today’s higher interest rates, these assets have lost significant market value, but since these losses have not been realized, they are not reflected in bank asset valuations or regulatory capital measures."
I don't read many econ or finance blogs at all - being over my head - but I feel like I read some pushback against this contention [e.g. B of A holding company: 0 capital] somewhere in the past year ... Is this situation regarding securities, loans really as straightforwardly dire as presented?
What is "A16Z"?
https://a16z.com/the-little-tech-agenda/
What is an "a16z startup"? Is that somehow different than other startups or other tech startups?
VC firm. a16z.com
First, thank you for your work on this issue.
I think it is worth also noting that the Dickey amendment effectively outlawed federal funding for firearm-related research from 1997 to 2013, with small funding until 2020, more after, and lifting of restrictions just this year.
Unfortunately, I'm not hopeful lifting of the ban will result in any funding of research on gun benefits. It would be great if there could be a relatively unbiased look at injury prevention that considers defensive use, accidents and violent offenders (crimes of passion and habitual criminals) while minimally interfering with recreational use.
My browser (or the website, or something else ?) won't register my "Like" for this comment. But rest assured, Guest, I do like it. I'm convinced that legal gun ownership poses a significant deterrent to crime.