Another term used is identifiable (although learnable and identifiable are not synonyms, I think identifiability is one precondition for learnability).
Identifiability means that out of all possible models, you can learn the correct one given enough samples.causal identifiability has some other connotations
See here https://causalai.net/r80.pdf as a good start (a nose in a causal graph is Markov given its parents, and a k-step Markov chain is a k-layer causal dag)
I see this argument often but for me it misses something.
The difference is about power. The wealth being this concentrated means the power is concentrated.
If people are okay with the idea of an ETF, or a wealth manager (or any type do fund manager/investment bank) then they should be okay with sovereign wealth funds/national ETFs that provide dividends with a guaranteed single share single vote setup.
If you want competition, then the US government used to be good at creating and sustaining artificial compétition in military procurement - similar to how Amazon let's teams compete on the same projects internally.
Because the competition would be artificially and enforced by laws, there's just as much as potential for massive efficiency gains as there is potential for corruption (the Norwegian national wealth fund has gone swimmingly for them)
You need to think about 1) the latent state 2) the fact that part of the model is post trained to bias the MC towards abiding by the query in the sense of the reward.
A way to look at it is that you effectively have 2 model "heads" inside the LLM, one which generates, one which biases/steers.
The MCMC is initialised based on your prompt, the generator part samples from the language distribution it has learned, while the sharpening/filtering part biases towards stuff that would be likely to have this MCMC give high rewards in the end. So the model regurgitates all the context that is deemed possibly relevant based on traces from the training data (including "tool use", which then injects additional context) and all those tokens shift the latent state into something that is more and more typical of your query.
Importantly, attention acts as a Selector and has multiple heads, and these specialize, so (simplified) one head can maintain focus on your query and "judge" the latent state, while the rest can follow that Markov chain until some subset of the generated+tool injected tokens give enough signal to the "answer now" gate that the middle flips into "summarizing" mode, which then uses the latent state of all of those tokens to actually generate the answer.
So you very much can think of it as sampling repeatedly from an MCMC using a bias, A learned stoping rule and then having a model creating the best possible combination of the traces, except that all this machinery is encoded in the same model weights that get to reuse features between another, for all the benefits and drawbacks that yields.
There was a paper close when OF became a thing that showed that instead of doing CoT, you could just spend that token budget on K parallel shorter queries (by injecting sth. Like "ok, to summarize" and "actually" to force completion ) and pick the best one/majority vote. Since then RLHF has made longer traces more in distribution (although there's another paper that showed as of early 2025 you were trading reduced variance and peak performance as well as loss of edge cases for higher performance on common cases , although this might be ameliorated by now) but that's about the way it broke down 2024-2025
I'd encourage everyone to learn about Metropolis Hastings Markov chain monte carlo and then squint at lmms, think about what token by token generation of the long rollouts maps to in that framework and consider that you can think of the stop token as a learned stopping criterion accepting (a substring of) the output
Without armchair parenting: people always say this, but the better parents in my life I have have never ever seen using punishment, only consequences that were clearly communicated ahead of time. And for those parents, looking disappointed or hurt seems to cut their kids like knives without any further punishment needed.
No it's not. If you run a trust based economy without clear property rights, this whole scenario doesn't exist. If you have property rights, you have a state to enforce them, so that state has a court, sues Alice (or Bob), sues John, the state punishes John, done. If you try to establish property rights without a state, then ans only then does it become interesting, but why would you do that? Establishing a democratic "state" is so simple and cheap that people do it all the time (associations, shareholder agreements), and in the end you need to physically enforce the property rights anyway, so why bother with the whole decentralised, anonymous rigamarole?
No, you aren't going insane. Cryptocurrency (block Chain ledger+distributed consensus) so far has no use case that is not covered by better technology. The one thing it facilitates (an online, permissionless, trustless, decentralised,ledger) has no real use case in human society, because in order to run the infrastructure to go online you already need to solve a lot of not-on-chain problems, and we have solved them with the technology of trust facilitation: the rule of law, legitimate authority, checks and balances etc. are all social technologies which are not perfect but have been honed over centuries and when combined with the internet to augment human discussion and decision making are incredibly efficient and low cost. It is pure insanity to try and replace this with cryptocurrency.
What makes sense is for governments and other societal trust anchors to run federated ledgers with cheap APIs, that everyone involved take local backups of and can verify, but go through specialised adapted write APIs (i.e. consensus through regulations, governance, checks and balances etc). Because that is useful and can automate notaries for example.
Everything else so far is either a decentralised Ponzi scheme (no clear owner/central figure, everyone can profit by luring in more suckers, but unlike the trust based economy no actual value is produced and the system would collapse if the trust based economy shut down the inflow of new capital - while even Venezuela and north Korea still produce things under heavy sanctions and could maybe be fine if they got rid of their cleptocrats and went back to an agrarian society, albeit with a lower QOL than the rest of the world) and/or derives it's value from using the Ponzi driven price of the currencies to launder money or provide tools to speculate/keep the game going. It's all recursive and contingent on going to the moon, with no there there. The epitome of default dead.
The market can stay irrational longer than we can bet against it. It is fully possible (and I'm afraid of it happening) that crypto bros get enough lobbying power to regulatory capture and insert themselves as "decentralised middlemen" into governmental processes. If this happens, they might point to the "value added" by Cryptos in the same way TurboTax in the US currently claims its usefulness. Bug we should not believe these lies, cryptocurrency is pure rent seeking.
That's why you're not going insane, nor missing something. There is simply no there there, just a bubble big enough to capture the very state it claims to dislike while it uses energy and IT infrastructure only that state and its society enabled
Everyone might own cars, but if one population can only afford them if they do their own work on them, such a ban overly impacts that group.
"AH", you might say, "but that is just discrimination against poor people, there are poor whites as well!". Yes. But if you look at the total wealth distribution in the US, last time I checked non-whites are much more likely to be on the bottom tiers than whites, given their skin color and nothing else. So the class issue is also a (systemic) racism issue.
"Sounds like poor whites and non-whites are natural political allies" you might say. Yes actually. But that does not erase the racism issue, since the different parties in that political coalition will have different needs and interests which need to to treated with respect by all members in order to give a lasting coalition.