vladimirralev

It depends. Some banks will ask for root origin of funds. Say you earned some money 20 years ago, invested them with various brokers into stocks all over the world, real estate etc. And then cashed a large sum into your account. If any one of those transactions is flagged they will require you to go back 20 years and collect those invoices that you earned and then you have to collect every transaction from the brokers, this may involve foreign jurisdictions where record keeping and querying is not available from 20 years ago.

So you will fail the disclosure and the bank will flag you in the shared blacklist DBs, they close your accounts and you are out of the system. This is happening enmasse to people who invested in and cashed out of real estate abroad(asia/africa/sa) some time ago. I imagine it's the same with foreign stock markets. And you never know what will be required 20 years in the future, thus most financial advisors discourage "moving money around". They even have a whitelist of brokers, stocks and jurisdictions, anything outside that is uninvestable for normal people just because of KYC/AML paperwork requirements.

That's what I mean. The Jensen inequality applies to any distribution, but the intuition presented in the post is only good for simple distributions and all examples are gaussian/binomial-like. It would be difficult to raise the same points with something multimodal or arbitrary.

Looks like a reasonable intuition guide with a couple of caveats. This intuition only works for gaussian or may be a few other distributions, not for a general p(x). Then the EV rows in the tables are not actual EV quantities, only the total is an EV, that can confuse somebody. Overall I think, the point could have been carried better with a few nice charts showing the p(x), f(p(x)) and the E.

One can make the case for a perfectly rational adversary who always avoids picking paying numbers in anticipation of the opponent to exclude paying numbers successively in their guesses. When the game is played with perfectly rational characters the picker is doomed to select one specific number and thus you always make the maximum amount. There are some variations of the binary search but that can also be worked around. If they are not cheating, that is.

I have the same experience. I have my own benchmark, I take a relatively complex project like FreeSWITCH on github, which is part of the training set for all coding LLMs anyway so they should know it and I ask the AI to code small snippets, tests, suggestions and fixes to see how well it understands the codebase and the architecture.

I just tried the latest Cursor + Sonnet and it failed in every task. The problem is that there is no way to understand the code without either complete understanding of the domain and the intents or running it in some context.

Telecom and media domains in particular are well documented in specs and studied in forum discussions. I am sure they are part of the training data, because I can get most answers if asked directly. So far the LLMs fail to reason about anything useful for me.

Shockingly, there is a related algorithm that is widely considered much easier to understand and implement. https://cp-algorithms.com/string/z-function.html

And I can certainly implement the Z algorithm in a few mins while I struggle to implement the KMP off the top of my head.

Edit: perhaps a bit better article on this https://www.geeksforgeeks.org/z-algorithm-linear-time-patter...

The VP9 codec I imagine would heavily benefit from SIMD? Just SIMD should massively outperform any advantage gained by putting this into the kernel. Why does the kernel need an unoptimized codec?

First, don't count on AI being aligned at all. States who are behind in the AI race will increasingly take more and more risks with alignment to catch up. Without a doubt, one if the first use cases of the AI will be as a cyberweapon to hack and disrupt critical systems. If you are in a race to achieve that alignment will be very narrow to begin with.

Regarding the pet vs humans - the main difference is really that the humans are capable of understanding and communicating the long term consequences of AI and unchecked power, which makes them a threat, so it's not a big leap to see where this is heading.

> With a fully automated economy, all 8 billion of us can have that

We probably can't. I mean why stop at humans? Let's give every pet the same luxury, or ... in the limit we could give this to every living being. Ultimately someone is going to draw the line who gets what and who is useful or not "for the greater good".

It just happens that many living beings don't contribute to the goals of whoever is in charge and if they get in the way or cause resource waste nobody will care about them, humans or not.

Human rights and democracy is all cool, but I think we just witnessed enough workarounds that render human rights and democracy pretty much null and void.

The state space here is subsets of V, but not all possible subsets and not exponential. The entire state is represented by the priority queue at each step which is bounded by the number of nodes and edges. And each state represented (encoded) by the priority queue transitions to exactly one new state (by popping the top node from the queue and adding the neighbours). And the number of steps is also bounded by the number of edges and nodes in V.

Most importantly, at every step of the algorithm, we only need to preserve the previous node in the state space (the previous state of the priority queue) to compute the new state (pop the top and add the dirty neighbours) and we can forget all previous states that existed before. We don't even need to keep track of the visited nodes, although if we did, that wouldn't change anything.

Dijkstra is definitely dynamic programming, no doubt about it, you are still computing a new state from neighbouring state nodes while ignoring substantial number of non-neighbouring states.

You just have to accept the more abstract definition of "state" where the state encodes subsets of the graph. The states in Dijkstra are represented by subgraphs and distances that incrementally include more nodes leaving us with a path of states to follow in some order.

It's not much different from the traveling salesman or viterbi in that sense. You come up with some topological order of states in abstract state-space and then follow that topological order to compute the new state based on the adjacent previously computed states and never look back to update already finalised states.

With this more abstract point of view, it's clear Dijsktra is dynamic programming.

There is a whole field of graph dynamic programming problems. And the closely related Markov chain problems.

My advise is don't take Erlang's fault-tolerance promises too seriously. It's just a little framework that helps in some cases and gets in the way in other cases.

I've seen many Erlang systems fail in funny ways, including some of the big examples given here. Supervision trees are cool but it's clearly nonsense to hardcode restart strategy and timing numbers for workers as if all failure modes are the same and deployed in the same network/capacity/resource/conditions with any number of workers. The strategy and schedule for recovering 10 crashed resource workers will clearly be different when you have 1M workers. The strategy will be different if you are timing out on network or if you are getting a resource error and have better things to do than restarting workers.

Focus on fault-tolerance outside erlang - have standby capacity in isolation and load-balance properly, shard the system in isolated pieces as much as you can.

I keep saying Markov process because it encodes any kind of chance over time on a large scale. It encodes how smart the entrepreneurs and investors are born, how many opportunities they get per time step, how corrupt the regulators are, how the competition will interfere. There are real averages in numerical terms for all of these parameters that can be used. A coin flip can represent an arbitrary model if you tweak the time lags and the payouts. And if the game is large enough you can fix the payouts and extend the lags until you hit your desired fixed payouts. Coin flipping on a large scale can capture almost any economic model.

Well, no. This game sucks for individuals. That's not the point though. The point is that even though the game sucks, this is the game that represents the economic reality. And while this tweak of you or somebody deciding when to pay is a real part of the game, it is still a markov process that will result in the same wealth distribution due to the law of large numbers and people settling at the endpoints all the time, no matter how delayed in time or how small the chance.

True, but with some tweaks this is a representative model.

You can just carry any non-100% gain/loss "swimlane" as a game still-in-progress. Eventually on larger time scale, puchasing power adjusted or not, a swimlane of an investment will certainly hit one of those outcomes.

The outcome is a function of time but a large number of swimlanes guarantees that you can sample any time and observe the same distribution of settled outcomes. You can reasonably assume that the probability of transition from unsettled to settled state at any time is constant. You could model it as markov process. In fact there are many economic agent models doing something like this (OLG models where generations of economic agents overlap in time as child/worker/retiree). Even though these models make sense only on generational scale, with enough agents you can see the micro-structure being consistent in smaller time scales by "ticking" or sampling at smaller time steps the markov process.

The spending power issue could be ignored as "without loss of generality" in most cases or the thesis could be restated in terms of something that preserves purchasing power as part of the coin toss game.

Investment returns mean-revert only if allowed to by policy. Policy can override economic forces for very long generational time scales, thus making it still relevant for individuals.

And I personally don't agree with (or don't understand) these models being used for micro-structure analysis, but I think macro-level interpretations such as wealth distribution may be valid.

It's pretty disgusting, but we all come from a long history of violence and pain. The only chance these species have to survive long terms is for us humans to figure out how (preserve DNA, leave earth, simulated environments, and so on).

In that sense, I wonder if it may be fair to preserve the DNA of victimised animals and clone them somewhere in the future when resources allow, to have a peaceful life.

Not aware of this type of simplistic embeddings. I think getting the first few layers of a large pretrained vision model will get you better results. Blindly learning from a generic image would probably steer it toward the dominant textures and shapes rather than linear operation of camera moves.

The simplest embeddings for vision should focus on camera primitives and invariants. Translation, rotation, scale, skew, projections, lighting. It doesn't matter that much what you use in the layers, but you should steer the training with augmented data. Like rotate and skew the objects in the batches to make sure the layers are invariant to these things.

Next are some depth-mapping embeddings which go beyond flat camera awareness.

The best papers I've seen are face embeddings. You can get useful results with smaller models. There are of course deeper embeddings that focus on the whole scene and depth maps but those are huge.

In naive ML scenarios you are right. You can think of JPEG as an input embedding. One of many. The JPEG/spectral embedding is useful because it already provides miniature variational encoding that "makes sense" in terms of translation, sharpness, color, scale and texture.

But with clever ML you can design better variational characteristics such as rotation or nonlinear thing like faces, fingers, projections and abstract objects.

Further JPEG encoding/decoding will be an obstacle for many architectures that require gradients going back and forth between pixel space and JPEG in order to do evaluation steps and loss functions based on the pixel space (which would be superior). Not to mention if you need human feedback in generative scenarios to retouch the output and run training steps on the changed pixels.

And finally, there are already picture and video embeddings that are gradient-friendly and reusable.

The economy is just increasingly fake or rewarding foolish scummy behaviour instead of honest cooperative work. People coasted for 10+ years now on "passive income" from printed money along with almost no enforcement action against scams. That gave rise to a chain reaction of people trying to scam each other in a general revenge.

Take for example a big slice of the economy - real estate. Honest real estate agents who try to match real needs to real properties were totally displaced (or outcompeted) by agents who spam fake ads in order to secure an exclusive contract with unfair terms and sell you the biggest property you can afford with no regard of your actual needs. The Fed in turn backstops any losses and buys toxic mortgages perpetuating the scheme.

Further, imagine an economy entirely comprised of real estate agents (or even NFT traders). In such economy, there is only so much real need for agents, at some point there will be more agents than the market really needs. At that point the only way to survive in the market is to become adversarial, it is in your best interest to outspend everybody on deceptive marketing, then mismatch buyers and sellers so that they will have to come back to you looking for new deals once they realise they were mismatched. And again the only reason people can keep this mismatched economy going is because the Fed/congress flood the system with money and their mistakes are rewarded (in asset price inflation at least).

This is the economy we have now more or less in the different sectors. And policy makers can't touch anything because they prioritise stability and growth.

With handouts, you only redistribute the wealth for an instant and you need to keep printing to sustain it. On top of that, the stability of this system becomes dependent on active approach of handing out money (i.e. tight dependency on central government).

And I am not saying not to help the poor people. I am saying printing money doesn't help poor people, it helps the rich. If you want to help poor people like I said you can tax the excess, or you can provide a specific minimal food/energy/shelter. But again, handing out money for free only helps the top of top.

As a side note, you know a lot of the poorest people in EM and frontier markets save money in USD either to escape their own inflating currency or as a dollar peg in their country. These people are poor and suffer the inflation in global markets, but they do not benefit from the US handouts. They can never outbid a money printer. And that's in fact the likely cause of the riots across the world in countries that don't have swap lines with the Fed.

If you take it just one step further, the people "that need it most" really only need it in order to instantly donate it to the megacorps with pricing power at ever-increasing prices. The handouts help only in that instant and at the very next clock cycle that money is now in the hands of the megacorps. Printing money and handouts only help the top of the top in the pricing power hierarchy. If you want to actually help people you'd be doing taxes on excess. Everything else ends up in gigantic accumulation of wealth at the top in just a few steps down the game.

The only reason for people to take money out of circulation is because nobody managed to convince them to spend or invest in something worthy. If the economy can't produce enough worthy money sinks, that is a structural problem with regulation, entrepreneur skills/education and general business UX.

Printing money doesn't fix any of these issues, the only thing money printing does is force the money into the overcrowded money sinks (AAPL) and scams (NFTs). And I am willing to argue this makes the problem worse, because talent is now wasted into supporting scams or the best talent is retiring on passive income from money printing instead of finding new ways to convince investors and consumers to spend productively.

I know this is the mainstream, but all of this reads like total nonsense to me.

1. Price changes carry important information too. What if demand stays the same but price still goes up? It could there is too much money, or supply is constrained or a new tax was imposed in the chain, or something was banned or the market is simply inefficient.

2. Low rates are sign that money has been tight? Maybe in a world where rates are determined by a market, but right now the rates are whatever the Fed commands. If the Fed wasn't suppressing rates, all rates would be much higher. Further the demand for money is always infinite. Ask anybody on the street if they demand more money and 100% will say yes. Why doesn't the Fed increase the money supply for the random Joe on the street, but does so to cover unsustainable liquidity commitments made by banks?

3. Interest rates certainly drive relative prices as they widen and shrink the gaps between those on fixed income and everybody else. As such the different cohorts evolve different utility functions and preferences. For example certain asset prices can go higher than the price of food from interest alone.

4. Both inflation and NGDP are broad aggregates that clump together the top 1% with the rest with no regard of inequality, climate change, social unrest. It's easy to slip into a mode where the top 1% does extremely well while 99% are left behind and rioting, all satisfying the inflation or NGDP target. A mockery of a feudal economy in a way. And this is where we are heading.

5. We can prove this one as being false. Money neutrality has never been demonstrated in the long term. In fact most forms of money ended up guided by politics, hyper-inflated the supply, was banned, price-controlled or otherwise collapsed due to some inevitable populist political decision.

It's pretty funny how the Fed keeps telling us they look long term, inflation is transitory and markets will work it out in a few years without intervening. But when the repo market crashed they didn't wait "for the market to work itself out", they rewrote their whole rulebook and launched support facilities the same day.

By now it's clear they are making it up as they go and they are just dominating the system with increasingly excessive interventions into a self-exciting oscillation. The Fed increasingly need to resort to breaking the law to achieve their goals. The Fed is only allowed to buy government-guaranteed assets with the full faith that they will be repaid with taxes. They are not supposed to buy mortgages, ETFs, junk bonds, muni bonds. They've had these discussions in the past and determined it's against the federal reserve act. There is no doubt that the authors of the act never intended for this to be possible and the states wouldn't have even signed on the act if it was presented to them in such form.

Again, how do you define a shortage? Chip production is way above the 2019 levels. Moreover the utilisation of the fabs, new and old processes combined is much higher[1]. Semiconductor fab worker count is at all time high, 10% above 2019. Why isn't this excess demand (caused by too much free money) as opposed to a "shortage"?

[1] https://www.semiconductors.org/chipmakers-are-ramping-up-pro...

I really want an opportunity to unpack this with a real economist, because it doesn't make any sense. How do you define a short supply of goods? We are far above the 2019 supply of goods in nominal terms, in real terms, in terms of shipping containers and actual ships and trucks moving. At what point it stops being a supply issue and becomes an issue of excess demand? And wouldn't this demand be fuelled further by the fear of goods costing more tomorrow than they cost today, again due to the accelerating inflation.

Shelter is 33% of the basket weight. Gasoline is like 1%.[1] This is not an energy problem. This is just textbook populist dumb inflation from printing too much money and handing it out for no work/productivity. [1]https://www.bls.gov/cpi/tables/relative-importance/2020.htm

I think this article is overlooking the most important reason for this. Most senior developers have plenty of assets and are retiring or taking a long vacation with stock/crypto/real-estate gains from the past year.

Some are rolling out their own businesses sure, but it's a terrible time to start a business now. Wages are skyrocketing, supply chains are overbooked, inflation makes it impossible to secure long contracts on predictable terms, and the passive income mania fuelled by the banks makes working on fixed income unappealing.

It is better than coin toss. If you can't communicate a convincing solution or approach to your future team mates why should they hire you?

I think the greatest damage was done in the state of mind of regular workers. Nobody wants to work for cash that depreciates that fast. While passive asset holders are seeing gains exceeding normal wages for no work at all.

Watching bankers and policymakers ignore this, everybody is gambling to try to get into the unproductive game of sitting and waiting central banks to inflate their asset prices and retire as early as possible.

Inevitably, you end up with no workforce and learned helplessness, waiting for the government to step in and guarantee jobs/UBI in exchange for everything else you got.

I wonder if it's possible that CureVac had to take some extra risks to be competitive this late in the race. With the Pfeizer/Moderna results already in, may be they aimed to deliver a lower dose vaccine with less side effects that would be more desirable, but that didn't pay off.