Quite a lot packed in here. Reading Hoffman's description of blitzscaling's benefits, it focuses in how it affects the competition. Who is DOGE's competition? Why, federal employees, of course. The other thing is blitzscaling is almost exactly how Agile programming is executed in the real world, not in its defined theoretical model. The supposed engine of the Agile model is the painstaking front-end requirements gathering, plus ongoing evaluation of the evolving product by end users. The 2 most expensive parts, and the first to get cut.
Anyone, anywhere harping on efficiency as a goal in itself is selling you a pig in a poke. The efficiency is almost always in the system itself, not the product or service delivered. As that idealized description of Palantir's business model describes, efficiency applied to an end goal is difficult, expensive, and political. Far easier to make the process more efficient, plus it increases profits! Collateral damage is just another process to manage, which can in turn be made more efficient, like Amazon delivery.
I suppose one could apply blitzscaling to the dissemination of memes, like religion, that go back thousands of years. Using force to convert populations to a particular religion might be counted as blitzscaling.
The Chinese social surveillance system is partially blitzscaling by technology, and partially human scaling of censors. The US is going for full technology, which I suspect is why errors keep happening, like relying on facial recognition by machines and not checking the results, sometimes with fatal consequences. I can easily imagine AI being embodied in some Omni Consumer Products ED209 by some tech company cosying up to Trump, just as I G Farben did with the Nazi administration with Zyklon B for use in the death camps.
Someone should rephrase Vance's disgusting verbiage about dealing with immigration into a pastiche of how Reinhard Heydrich talked about implementing the "final solution" to the Jewish problem. Vance has sidled up to becoming a neo-Nazi. Dorothy Thompson would undoubtedly have picked him out in her piece "<a href="https://harpers.org/archive/1941/08/who-goes-nazi/">Who Goes Nazi?</a>
Thank you, this explains why Silicon Valley has been so susceptible to the AI Magic Beans Scam: because it pretends to replace the biggest drag on blitzscaling, human labor, with fungible compute. You can blitz *anything* if you can paste "AI" in instead of people, and they're so desperate for it to be true that they're completely ignoring the fact that ... it .... doesn't ... work ...
This makes me wonder about the relationship of tech to overriding due process in *pre*-DOGE govtech. The use of Salesforce or Microsoft Azure among civil servants, e.g.
“All the above tends to systematically limit their accountability to workers and to others whose lives may be very substantially affected by their workings.”
I’ve been thinking about this type of thing a lot. Lack of accountability manifests in, as you say, a pile of shit that *someone else has to deal with*, and usually that means “every one of us”. Blitzscaling has demonstrated that there is almost no limit to the damage that power-hungry groups will try to impose upon the rest of society - internalize profits, externalize costs. It feels tragic and terribly hard to solve - since the shit atomizes into each person’s lungs, how can we measure the cost back to the actions and groups that caused the harm? It seems that the world is currently more complicated than many societies are willing to face, or capable of facing, so we keep letting the garbage pile get bigger. How long do we have until the commons are so toxic as to be uninhabitable?
A few things. (1) Instead of calling the unicorn-seeking strategy "power law" you could use Taleb's black swan. The shape of the distribution below the unicorns doesn't matter to this approach. (2) The one point Bracy misses, IMO, is the role of blitzscaling in identifying which firms do or don't have unicorn potential. Really, it's the old continuous improvement/management by stress idea applied to scale rather than speed or inventory control. By pushing firms to scale rapidly at all cost (yellow lights, even some red lights), you find out quickly who's going to drop out of the race. Then you can cut your investment and put it somewhere else. (3) Behind all of this, and I suppose right wing thought in general, is the presumption that profit corresponds to social value (benefits minus costs), so that hugely profitable businesses are justified by making huge social contributions. It's obvious why successful individuals and firms would think this, but it's especially ironic when applied to a strategy whose goal is to overwhelm competition. ("Competition is for losers.") Surely big profits for products with network externalities are about something more than just how much good you're doing for the world.
If the guy (Hoffman) who enabled the no real employees just contractors via a jobs scam website you have to be on to be employable in most cities is one of the 'good guys' we're really in trouble.
Maybe I'm missing something, but I don't know how you could ever work for a digital surveillance company started by Peter Thiel, Joe Londsdale, Alex Karp and claim that the current company has "pivoted".
That ex-Palantir guy was using classic consultant-speak. What he described is the proper and practical way to actually deliver what consultant sales reps promise. Navigating the cultural and political land mines in redesigning or replacing a business process is highly complex, time consuming, and difficult. Requires massive diplomatic skills as well as the ability to locate, extract and consume a lot of arcane technical data, which is often Larry over there who's the only person who knows how that part works, and for some reason has never documented it.
As far as working for Evil, Inc, most people do, but just don't get their nose rubbed in it at the office. Palantir probably isn't any different.
Re "power law logic", I enjoyed Shalizi's slide deck, but it applies to situations where you have a definite probability generating process but have mis-identified it as a power law distribution. Asking whether the SV growth model is "really" a power law seems like a wrong question; there is no definite underlying probability generating process which is "correct", the term is just being used informally to describe a preference for very fast growth. That's either more innocuous or even more annoying than looking for a straight line on a log-log plot, depending on your perspective.
But Levine's characterization "a basket of options is worth more than an option on a basket" seems fine.
Quite a lot packed in here. Reading Hoffman's description of blitzscaling's benefits, it focuses in how it affects the competition. Who is DOGE's competition? Why, federal employees, of course. The other thing is blitzscaling is almost exactly how Agile programming is executed in the real world, not in its defined theoretical model. The supposed engine of the Agile model is the painstaking front-end requirements gathering, plus ongoing evaluation of the evolving product by end users. The 2 most expensive parts, and the first to get cut.
Anyone, anywhere harping on efficiency as a goal in itself is selling you a pig in a poke. The efficiency is almost always in the system itself, not the product or service delivered. As that idealized description of Palantir's business model describes, efficiency applied to an end goal is difficult, expensive, and political. Far easier to make the process more efficient, plus it increases profits! Collateral damage is just another process to manage, which can in turn be made more efficient, like Amazon delivery.
In the case of DOGE however, collateral damage is "managed" by sweeping it under the rug.
I suppose one could apply blitzscaling to the dissemination of memes, like religion, that go back thousands of years. Using force to convert populations to a particular religion might be counted as blitzscaling.
The Chinese social surveillance system is partially blitzscaling by technology, and partially human scaling of censors. The US is going for full technology, which I suspect is why errors keep happening, like relying on facial recognition by machines and not checking the results, sometimes with fatal consequences. I can easily imagine AI being embodied in some Omni Consumer Products ED209 by some tech company cosying up to Trump, just as I G Farben did with the Nazi administration with Zyklon B for use in the death camps.
Someone should rephrase Vance's disgusting verbiage about dealing with immigration into a pastiche of how Reinhard Heydrich talked about implementing the "final solution" to the Jewish problem. Vance has sidled up to becoming a neo-Nazi. Dorothy Thompson would undoubtedly have picked him out in her piece "<a href="https://harpers.org/archive/1941/08/who-goes-nazi/">Who Goes Nazi?</a>
Thank you, this explains why Silicon Valley has been so susceptible to the AI Magic Beans Scam: because it pretends to replace the biggest drag on blitzscaling, human labor, with fungible compute. You can blitz *anything* if you can paste "AI" in instead of people, and they're so desperate for it to be true that they're completely ignoring the fact that ... it .... doesn't ... work ...
As I was reading this "Enshitification" came to mind.
This makes me wonder about the relationship of tech to overriding due process in *pre*-DOGE govtech. The use of Salesforce or Microsoft Azure among civil servants, e.g.
Wait a second. Marc Andreeson said that technology was the way to a better future and that faster was always better.
Now I am just confused.
“All the above tends to systematically limit their accountability to workers and to others whose lives may be very substantially affected by their workings.”
I’ve been thinking about this type of thing a lot. Lack of accountability manifests in, as you say, a pile of shit that *someone else has to deal with*, and usually that means “every one of us”. Blitzscaling has demonstrated that there is almost no limit to the damage that power-hungry groups will try to impose upon the rest of society - internalize profits, externalize costs. It feels tragic and terribly hard to solve - since the shit atomizes into each person’s lungs, how can we measure the cost back to the actions and groups that caused the harm? It seems that the world is currently more complicated than many societies are willing to face, or capable of facing, so we keep letting the garbage pile get bigger. How long do we have until the commons are so toxic as to be uninhabitable?
A few things. (1) Instead of calling the unicorn-seeking strategy "power law" you could use Taleb's black swan. The shape of the distribution below the unicorns doesn't matter to this approach. (2) The one point Bracy misses, IMO, is the role of blitzscaling in identifying which firms do or don't have unicorn potential. Really, it's the old continuous improvement/management by stress idea applied to scale rather than speed or inventory control. By pushing firms to scale rapidly at all cost (yellow lights, even some red lights), you find out quickly who's going to drop out of the race. Then you can cut your investment and put it somewhere else. (3) Behind all of this, and I suppose right wing thought in general, is the presumption that profit corresponds to social value (benefits minus costs), so that hugely profitable businesses are justified by making huge social contributions. It's obvious why successful individuals and firms would think this, but it's especially ironic when applied to a strategy whose goal is to overwhelm competition. ("Competition is for losers.") Surely big profits for products with network externalities are about something more than just how much good you're doing for the world.
If the guy (Hoffman) who enabled the no real employees just contractors via a jobs scam website you have to be on to be employable in most cities is one of the 'good guys' we're really in trouble.
Narrator: they were really in trouble.
Maybe I'm missing something, but I don't know how you could ever work for a digital surveillance company started by Peter Thiel, Joe Londsdale, Alex Karp and claim that the current company has "pivoted".
That ex-Palantir guy was using classic consultant-speak. What he described is the proper and practical way to actually deliver what consultant sales reps promise. Navigating the cultural and political land mines in redesigning or replacing a business process is highly complex, time consuming, and difficult. Requires massive diplomatic skills as well as the ability to locate, extract and consume a lot of arcane technical data, which is often Larry over there who's the only person who knows how that part works, and for some reason has never documented it.
As far as working for Evil, Inc, most people do, but just don't get their nose rubbed in it at the office. Palantir probably isn't any different.
1. Larry isn’t paid to document.
2. Larry is too busy to document.
3. Larry doesn’t like to document.
You can argue about the end products and what purpose is served, but metastatic cancer is certainly optimized for growth.
We've seen this show before and it is called "Springtime for Psychopaths."
"Once the rockets are up, who cares where they come down?
That's not my department, says Wernher von Braun."
"Smart" is a funny term to describe people who use lots of words to defend the indefensible.
If the Intelligence Quotient exceeds the Humanity Quotient there is gonna be trouble.
Blitzscaling is another word for being last firm left standing after bankrupting competitors. As a strategy, outspending peers fails (https://qz.com/1540608/the-problem-with-silicon-valleys-obsession-with-blitzscaling-growth) during down-turns or the cost of capital is non-zero (no quantitative easing). Specific failures
- short/long spread - WeWork long-term leases and short-duration rentals
- bad unit cost economics - eg Pets.com
- premature optimisation - Solydyne - picking less scalable solar tech
If you're betting on a spread of portfolio it might be justified but the numbers suck for individual founders.
fascinating insights. thank you.
Re "power law logic", I enjoyed Shalizi's slide deck, but it applies to situations where you have a definite probability generating process but have mis-identified it as a power law distribution. Asking whether the SV growth model is "really" a power law seems like a wrong question; there is no definite underlying probability generating process which is "correct", the term is just being used informally to describe a preference for very fast growth. That's either more innocuous or even more annoying than looking for a straight line on a log-log plot, depending on your perspective.
But Levine's characterization "a basket of options is worth more than an option on a basket" seems fine.