Posts from arxiv.org
All PoW/PoS coins are screwed in the long termYes, a rather callous title, in the hopes that people will come in here to tell me why I'm wrong. See the bottom of this post for a TL;DR. My thesis is that cryptocurrencies relying either on PoW or PoS, cryptocurrencies with inflation, fees & staking, cryptocurrencies with block subsidies and reward schedules are all screwed in the long run. My reasoning for this is that cryptocurrencies using PoW, PoS, or anything like it, actively undermine their own goals by incentivizing centralization over time at their core. In doing so, these protocols encourage a loss in stall resistance and a loss in security. I also argue that at least 2 cryptocurrencies (IOTA and Nano) solve this issue through their feeless/inflation-free proposition. Why Bitcoin is screwed Bitcoin mining offers rewards. These rewards consist of a block subsidy (money supply increase, currently 6.25 BTC per block) and fees. These rewards (mostly) go to those with the highest hash power. Bitcoin mining is a business. It's a business focused on cost efficiency, because the revenue side is largely unchangeable by miners. Total costs consist of energy costs, ASIC purchases/writedowns, capital costs, rent of the location, maintenance, etc. Almost all these costs have economies of scale associated with them. If I'm a large miner, I have a stronger negotiating position for ASICs. I have a stronger negotiating position for energy contracts. I have access to cheaper capital, I can more efficiently maintain my ASICs. Combine mining rewards with economies of scale for mining, and what you get is centralization over time. The largest miners have the lowest cost-base, making the most profit, being able to reinvest more in ASICs, increasing their share of consensus over time. This isn't some radical, unsupported take. The theory is quite clear, and is why we tend to have anti-trust legislation in most countries. Research also backs this up, I'll link to some papers on it at the bottom of this post. FUD, China is banning mining so miners will disperse more broadly, we have Stratum V2 coming, miners will join different mining pools, nodes are the ones that matter not miners, we don't see 80% belonging to one miner now! None of the above changes the centralization in consensus power over time. It doesn't change the economic rationale. China banning mining means there isless dispersion, as there are now fewer locations where mining is possible. Stratum doesn't fix the incentives. Miners can join different mining pools (though history shows they don't ) but it's about the underlying miners, not the mining pools. Not to mention that mining pools themselves are far more centralized than most people think (see 3) in the links below). Nodes can check the chain all they want, those with the consensus power decide whether to include transactions. If I had a majority of mining power, I wouldn't outright show it. I would send in increasingly higher fee transactions, forcing people to pay a lot for me to process their transaction. Unbelievable? CheckMiner Collusion and the Bitcoin Protocol to see that hundreds of millions in excess fees are already being paid. Good thing I'm not in Bitcoin but in -insert other PoW coin here-. The incentives and trend aren't different for other PoW coins. It's just less visible as Bitcoin has a larger market cap, so the incentives are biggest here. Mining is terrible for environment anyway. Good thing I'm in PoS coins! Right. Without economies of scale in consensus, PoS is immune from this centralisation over time, right? No, and this series of steps should be even easier to follow than that for Bitcoin. When you stake the most coins, you get the most rewards. Those that get the most rewards grow fastest. In many PoS cryptocurrencies you need a minimum amount to stake in the first place. As a regular user using the network, you might not want to lock up your stake but rather use your coins to transact, paying fees while doing so. Some cryptocurrencies try to make the network seem more decentralized through maximizing the size of a single pool, which is a bit like saying that we can increase Bitcoin's decentralization by splitting AntPool into Ant and Pool. Nothing has changed, if anything it's simply muddying the waters by obscuring how centralized the system really is. All this might not matter much to those in crypto for trading/short term gains. However, the literal defining property of cryptocurrency is being decentralized. It's the mechanism to ensure security, it's what provides the underlying value in the store of value narrative for Bitcoin. It's why we are okay with sacrificing some performance relative to centralized payment processors/apps. By becoming ever more centralized over time, cryptocurrencies' security and underlying value is decreasing over time, rather than increasing. Possible solutions The common thread in both PoS and PoW is that there are mining rewards. These rewards are offered in compensation for investing in hash power, for locking up a stake, for securing the network. It's the incentive that's needed to make people spend money, render their coins less usable, or otherwise take some form of risk. The simplest solution then is to remove these mining rewards. Remove block subsidies, remove fees, and there is no centralization over time inherent in the protocol as the big do not get bigger. As far as I know, only two major cryptocurrencies are both feeless and inflation-free: Nano and IOTA. Both chains rely on other incentives for transaction validation. In Nano's case, the theory is that wanting trustless access to the network and deriving value from the networkincentivises people and businesses to run validators . In IOTA's case, the incentive is that by validating others' transactions, you give yourself the option to transact.See here for a longer take. Does this have trade-offs? In both IOTA and Nano's case, the feeless proposition meant needing to look for a different transaction prioritization and anti-spam mechanism. In both cases, a small (tiny, rather) PoW is needed to create a transaction. In IOTA, prioritization under congestion is done through mana, which can be rented . In Nano, since recently prioritization is done through a combination of account balance and time since last transaction . It needs to be said that this IOTA implementation is still mostly theoretical on mainnet. They've had trouble the past years actually getting IOTA working without a central coordinator (making IOTA's mainnet centralized for value transfers), because the Tangle that IOTA uses is notoriously complicated and difficult. The IOTA Foundation claims to have found the solution now. As someone who has been following IOTA for a while and gotten burned during that time by believing the timelines they announced, I take a wait and see approach here. That being said, the lack of centralization over time is clear. In Nano, a recent spam attack lead to issues following which the aforementioned prioritization by account balance and time since last transaction began to be implemented. However, Nano's proven to be able to handlemillions of transactions per day on its mainnet. More importantly, having had a decentralized mainnet for years, Nano is proving more than any other cryptocurrency that it is possible to have a decentralized cryptocurrency without fees and without inflation with high security. Over the course of ~120 million transactions, Nano has never had a doublespend nor chain re-org, something many other cryptocurrencies can't say. Over the course of these years, there have consistently been many validators running, validating the theory that without fees and inflation, there is enough reason to run validators. Without mining and without staking in Nano, centralization over time is absent from Nano at a core level, leading me to believe that unlike 99% of cryptocurrencies it's not screwed in the long run. For more information on the design and consensus of Nano,see also this article . Making a long story short Every cryptocurrency that has fees and/or inflation has a trend towards consensus centralization over time. This centralization degrades the security and underlying value of a decentralized network over time. This may not be obvious yet, but without countervailing forces there is no reason to believe this trend will reverse over time. Feeless cryptocurrencies like IOTA (theoretically) and Nano (in practice) solve this through a lack of mining rewards. I believe this is the best (only?) way to ensure true decentralization in the long term, and believe that true to the title of this post, cryptocurrencies that centralize over time are screwed in the long term. I'd love to hear what PoS/PoW coin supporters think of this, and where the mistakes in my reasoning are. If there are other cryptocurrencies that are also feeless/inflation-free, I'd love to hear so too. * Trend of centralization in Bitcoin's distributed network . * Decentralization in Bitcoin and Ethereum Networks . * A Deep Dive into Bitcoin Mining Pools . * Centralisation in Bitcoin Mining: A Data-Driven Investigation . * Miner Collusion and the Bitcoin Protocol .
Why is AI hard and Physics simple?We discuss why AI is hard and why physics is simple. We discuss how physical intuition and the approach of theoretical physics can be brought to bear on the field of artificial intelligence and specifically machine learning. We suggest that the underlying project of machine learning and the underlying project of physics are strongly coupled through the principle of sparsity, and we call upon...1
Regularization is all you Need: Simple Neural Nets can Excel on Tabular DataTabular datasets are the last "unconquered castle" for deep learning, with traditional ML methods like Gradient-Boosted Decision Trees still performing strongly even against recent specialized neural architectures. In this paper, we hypothesize that the key to boosting the performance of neural networks lies in rethinking the joint and simultaneous application of a large set of...1
The incel lexicon: Deciphering the emergent cryptolect of a global misogynistic communityEvolving out of a gender-neutral framing of an involuntary celibate identity, the concept of `incels' has come to refer to an online community of men who bear antipathy towards themselves, women, and society-at-large for their perceived inability to find and maintain sexual relationships. By exploring incel language use on Reddit, a global online message board, we contextualize the incel...1
Surveilling Surveillance: Estimating the Prevalence of Surveillance Cameras with Street View DataThe use of video surveillance in public spaces -- both by government agencies and by private citizens -- has attracted considerable attention in recent years, particularly in light of rapid advances in face-recognition technology. But it has been difficult to systematically measure the prevalence and placement of cameras, hampering efforts to assess the implications of surveillance on privacy and...1
Thinking Like TransformersWhat is the computational model behind a Transformer? Where recurrent neural networks have direct parallels in finite state machines, allowing clear discussion and thought around architecture variants or trained models, Transformers have no such familiar parallel. In this paper we aim to change that, proposing a computational model for the transformer-encoder in the form of a programming...1
The Modern Mathematics of Deep LearningWe describe the new field of mathematical analysis of deep learning. This field emerged around a list of research questions that were not answered within the classical framework of learning theory. These questions concern: the outstanding generalization power of overparametrized neural networks, the role of depth in deep architectures, the apparent absence of the curse of dimensionality, the...1
Measuring the Algorithmic Efficiency of Neural NetworksThree factors drive the advance of AI: algorithmic innovation, data, and the amount of compute available for training. Algorithmic progress has traditionally been more difficult to quantify than compute and data. In this work, we argue that algorithmic progress has an aspect that is both straightforward to measure and interesting: reductions over time in the compute needed to reach past...1
Do the Angles of a Triangle Add up to 180°? -- Introducing Non-Euclidean GeometryHow can we convince students, who have mainly learned to follow given mathematical rules, that mathematics can also be fascinating, creative, and beautiful? In this paper I discuss different ways of introducing non-Euclidean geometry to students and the general public using different physical models, including chalksphere, crocheted hyperbolic surfaces, curved folding, and polygon tilings. Spherical geometry offers a simple yet surprising introduction to the topic, whereas hyperbolic geometry is an entirely new and exciting concept to most. Non-Euclidean geometry demonstrates how crafts and art can be used to make complex mathematical concepts more accessible, and how mathematics itself can be beautiful, not just useful.
Predictive Coding Can Do Exact Backpropagation on Any Neural NetworkIntersecting neuroscience and deep learning has brought benefits and developments to both fields for several decades, which help to both understand how learning works in the brain, and to achieve the state-of-the-art performances in different AI benchmarks. Backpropagation (BP) is the most widely adopted method for the training of artificial neural networks, which, however, is often criticized...1