Really great post from Modern Treasury. They are crushing it.
My only question with this explanation is -- to be a 3rd party service provider, this means they would need a somewhat intense bank integration with 1 side of the ACH, right? The retailers definitely just fill out an ACH release form like [1], and I really think the distributors don't do a heavy bank integration either...
Interesting... I don't know much about Fedwire, I will check it out. Thanks YuriNiyazov!
And yes sorry it's not free, just relatively cheap -- I think it's $200/year and $1 per transaction, which doesn't seem like enough to cover traditional wires.
Alright yuriniyazov, I’ve been reading for hours about Fedwire but I’m still a bit stumped. To participate directly in Fedwire you need to have a Master account with the Fed, in which case you need to basically be a bank. So they… have a secret subsidiary that’s a bank that has a fed master account? Or somehow a very, very good relationship with their bank such that their bank (which has a master account) proxies the Fedwire transactions for them at cost?
With the defaults of per_replica_batch=1, seq=2048 and gen_len=512, a completion takes about 20 seconds.
I'm not sure yet what settings I'll end up with if I decide to play with this more. per_replica_batch=3, seq=1024, gen_len=64 would give an experience roughly similar to the AI Dungeon that I'm used to, though less clever than the Dragon model, and a bit slower at about 10 seconds per batch.
This is awesome Kindiana! Two quick questions.
1) GPU: Any tips for converting this to run on GPUs (inference-only)?
2) Speed: Any advice for achieving faster speeds? 10-13s is a bit tough for commercial applications. And, would you expect a big performance hit as well from running on GPUs?
Honestly, I'm quite surprised by how much people here seem to think it's biological. It seems to me there is no way society's slight move toward encouraging gender-neutral playing isn't still strongly offset by the pervasive culture of machismo and femininity. The second your boy steps foot into school, if he's wearing something pink or playing with a doll, other boys will make fun of him. A single incident like that is a plenty strong incentive to stick with same-gendered toys.
And coming at it inversely, I've also seen some hypotheses that neurons are a bit more autonomous than many think, and that the brain could be seen as a complex colony of billions of single-celled organisms working together, with some analogies to aspects we see in large-scale human coordination problems and game theory.
Indeed. Business needs screen sharing, screen annotation and audio (Slack covers 99% of this). Home users are relatively happy with what they have already (although no 4K streaming yet and cameras in laptops are kinda tragic). That leaves porn.
Yup, porn is definitely adopt it as soon as it makes economical sense.
Especially, if this is pre-recorded and only sent for viewing (so no need too much bandwidth on upload side). I guess it would come down to the hardware and how much it costs.
“The total energy required to maintain a secure and stable fiat currency is likely orders of magnitude greater than that of Bitcoin today.”
No way. You hear this argument all the time — that fiat currency also uses tons of energy - but I just don’t think it’s true (relative to the amount of transaction throughout, and also in absolute terms). Bitcoin mining uses something like 0.5-0.6% of global electricity consumption for a tiny, tiny fraction of (purely speculative, not economically meaningful) transactions. What % do you think fiat uses?
That is one of the important points point made by jtoomin in the first post of this thread: the energy consumption of a PoW blockchain if not proportional to the number of transactions. This means we could have everyone, everywhere, using a PoW blockchain with the same percentage of energy consumption as today (although it would probably rise as price would arguably also rise as demand increases, but it would eventually find an equilibrium, it will never "consume all electricity").
Didn't bitcoin have the chance to double their transactions-per-second for a constant amount of power consumption, in the form of Segwit2x, and decide not to do so?
The fact Bitcoin could be more efficient and chooses not to isn't really an improvement over being unable to be more efficient IMHO - in fact arguably it's worse.
I completely agree, increasing the maximum block size is the simplest way to scale. Bitcoin BTC refused to do that, keeping the 1 MiB maximum block size instituted by Satoshi Nakamoto in 2010 as a spam-prevention measure, which led to raising transaction fees as users compete for the scarce block space. That is also the reason why Bitcoin Cash BCH was born in August 2017: to raise the maximum block size and allow the Bitcoin blockchain to process more transactions per second while keeping fees low.
At the technical level BCH has showed that the network is stable in production with 20 MiB blocks, but the goal and ongoing work is to support gigabyte and eventually terabyte blocks so it would scale to worldwide usage while keeping transaction fees low. The counterargument by Bitcoin BTC is that large blocks centralize the network, as not everyone can afford to run a full node. That's a different debate that I would be happy to discuss, but I want to start by pointing out that keeping the block size low also centralizes the network, as many users will not be able to afford the increasing transaction fees (which rose to $50 at recent times). Does it make sense to restrict the network so anyone can run a $50 node but not be able to transact on it? I think decentralization is a means to an end (electronic P2P cash), not a goal in itself. In other words, blockchains should be decentralized enough that no single entity can take control of the network, but too much decentralization is inefficient, so an equilibrium must be found.
> In other words, blockchains should be decentralized enough that no single entity can take control of the network, but too much decentralization is inefficient, so an equilibrium must be found.
I believe that is the hardest problem that bitcoin didn‘t solve. The hashrate is governed by the price and a arbitrary halving schedule. There is no mechanism to regulate the hashrate to a certain setpoint where attacking the network is not feasible. This issue becomes even more severe when the block reward is dominated by fees in the future. In that scenario the hashrate is not so much determined by the price anymore but the available block space and the utility of on-chain transactions. It is really hard to judge bitcoins security model considering all these variables. But for now the solution seems to be to vastly overcommit hashrate
It's not becoming more efficient if you are just allowing blocks to be filled with frivolous data that wouldn't have otherwise ever been put on the blockchain.
It's easy to increase the block size, but it's very unlikely it would ever be decreased again, so it makes sense to wait until there is no other option.
Bitcoin strength and value is on decentralization. If you increase the block size, it will be seized by governments or large corporations. This has long being debated, and the reason it is like it is today, is that everyone really vested into it agreed (in a distributed way) to keep it as so.
> If you increase the block size, it will be seized by governments or large corporations.
That doesn't make any sense. Any actor, like for example a government, might seize (control) a blockchain by having overwhelming mining majority (51% attack), but that has little to do with the size of the blockchain itself.
The only reason one might have to think that way is believing the false narrative that a BTC "full node" does anything to secure the network. Only mining nodes secure the network, having a copy of the full blockchain is not enough.
today with a raspberry pi and a 1TB hard drive you can have a node and be one of the validators. If hardware is unattainable by the users interested in helping with decentralization (and there are few enough of those) large corporations or governments will be the only validators and defeat the purpose of decentralization
This is only the case of you would arbitrarily increase the number of potential transactions per block.
But as far as I know the maximal number of transactions is capped in practice.
Furthermore the security of PoW is based on it being far more expensive to mess with it then to mine it. But if you push more mommy through the same number of blocks it's at some point breaking apart.
Lastly no one says it will ever consume all energy. That works be absurd. It's about the required energy (and hardware cost) in relationship to the provided value being unhinged and for security you need steadily increasing energy costs for increased usage.
> But as far as I know the maximal number of transactions is capped in practice.
What do you mean? There are no hard technical impediments to Bitcoin being able to scale to worldwide usage with basically the same structure as today. It's a matter of optimizing software and having powerful hardware (that is available today, and will be even cheaper in the future). This article by Johannes Vermorel [1] goes over it in detail.
If you have any evidence that goes counter to this I would love to read more about it.
