6,349 karma · joined May 14, 2017
Russia continued WW1 with a civil war that lasted until 1922. Their military deaths were quite evenly split between the world war stage and the civil war stage.
Logging a few hundred hectares was not a problem. Google could have done that without additional permits, if their intention was to plant a new forest afterwards. Environmental impact assessment was only necessary, because they intended to replace the forest with something else.
The key question is whether Google can comply with the law after the fact. If they complete the assessment and the government finds nothing that would prevent the construction of the data centers, the project can go on. Maybe there will be some minor sanctions, as well as some delays, as they didn't do things properly the first time. And even if there is something that prevents construction, Google can still probably comply by planting a new forest.
My guess is that the government will eventually find that all requirements have been satisfied and let the projects move on. But because this is now a controversial issue, someone with the standing will very likely file a complaint. Then the projects will be stuck in administrative courts for a year or two.
You can work around that to some extent with explicit instructions, but it doesn't scale nearly as well as actual model training.
In present-day Turkey, people typically have 5–15% Turkic ancestry. For the most part, their ancestors called themselves Roman, or Greek, or Lydian, or Phrygian, or Hittite, or whichever culture happened to be dominant at the time. When new conquerors came, they pillaged a bit and put their own people in charge. Most commoners, and even many cooperative elites, would then continue their lives as before. If the new elites managed to stay in power, people would often gradually adopt their language, religion, and cultural identity.
As for the Aztecs, I found it amusing that the previous Mexican ambassador to the US was Esteban Moctezuma. Guess from which ancestor that name came from.
Helsinki gets its water from the 120 km Päijänne Water Tunnel, which cost €200-250 million (adjusted for inflation) to build. Peak flow is 10 m^3/s by gravity alone and up to 20 m^3/s with pumps, which should be enough to cool a 1 GW data center. You could increase the cost by 2x (because real wages have gone up), by another 2x (because middlemen have destroyed productivity), and by yet another 2x (because everything is more expensive in America), and it would still be pretty reasonable.
I personally prefer going to restaurants when I'm in Europe and using delivery in the US.
The issue is tacit knowledge, or implicit context. Most details are never written down. If you don't know them from experience, you have to guess. If you guess, you often guess wrong. And even if you know something from experience, you are often not aware of it, until you see something that violates it. So it's not possible for you to write it down in advance.
You can replace people with AI in the above, and nothing fundamentally changes.
It's always easier to measure progress according to the internal metrics of the field than to evaluate the contributions to the wider understanding of the topic. In theoretical computer science (which I'm most familiar with), people have long complained about results focusing on shaving sublogarithmic factors from complexity bounds (while making the algorithm worse in practice) and about reviewers being impressed by the technical difficulty of proofs. But progress like that is easier to measure than new algorithmic ideas or conceptual understanding.
But it's not all bad. The researchers chasing the metrics are almost always genuinely interested in the topics they study. Their actual contributions mostly come from the ideas they explore while trying to achieve measurable progress. And because they are not expected to produce anything of direct value (Goodhart's Law for applied researchers), they can explore a wider range of ideas.
I personally noticed this when I moved from theoretical computer science to algorithmic bioinformatics. When I start a new project, the expectation is that researchers in genomics should be using sofware that uses the new algorithms five year from now. That expectation is useful, but it's also a strict constraint on what I can afford to try.
O-1 is also easier to get in fields like academia and entertainment, where everyone is building their individual brand. Corporate employees have a harder time qualifying, because their contributions are rarely attributed to them in public.
Those times are gone, but the theoretical justification that makes homes income-producting assets remains.
Gödel's incompleteness also constrains human and AI mathematicians. Both just strive to prove whatever can be proven in the system they are working in.
The actual conditional result was "if problem A is essentially this difficult, problem B is essentially that difficult". A specific formulation of it was proven, but it depended on a specific assumption that was just shown false. But the general result is still probably valid, because the general assumption holds. But there may be no point in going through the effort of proving another specific formulation, when the automatic theorem factory could just come up with another technical improvement targeting that formulation.
I started in theoretical computer science, but I quickly drifted to more applied areas, because I was annoyed with how often theoreticians would confuse the map for the territory. But if AI can now do that much more efficiently than any human, perhaps theoreticians will have to rethink how much they should focus on specific provable statements.
Conditional lower bounds are a way of building understanding of the essential difficulty of specific computational problems. But if AI can now routinely generate marginal improvements, conditional bounds based on unproven assumptions become a waste of effort.
This is mostly due to how mathematics works. Ideally, we would like to prove something like "if problem A is essentially this difficult, problem B is essentially that difficult". But what we actually prove is more like "if (specific formulation of the difficulty of problem A), then (specific formulation of the difficulty of problem B)".
But those specific formulations become fixed targets for the AI to attack. If it manages to break the specific assumption, for example by creating an O(n^1.9998) time algorithm that is for all intents and purposes worse than a naive O(n^2) time algorithm, the conditional result becomes void. We could try to salvage the result with a different formulation, but that again becomes a fixed target.
This is essentially Goodhart's Law. We measure improvement with highly precise metrics, while we are actually interested in qualitative understanding.
EDIT: If theorems and proofs become cheap, marginal improvements are no longer interesting. Qualitatively better algorithms or unconditional lower bounds would be actual contributions. As would be a specific formulation of a conditional result that is robust against technical improvements made by AI targeting that specific formulation.
There are a lot of results about conditional lower bounds: "If this problem is at least this hard, that other problem must be at least that hard." But now a widely used assumption was proven wrong, and an entire house of cards collapsed.
It feels like that particular research direction is now a dead end, until we can figure out a way of proving conditional bounds that is robust against technicalities. We would like to prove something like "If this problem is essentially at least this hard, that other problem must be essentially at least that hard." If the conditional bound depends on the assumption that the first problem requires at least n^2 time but somebody comes up with an O(n^1.9992) time algorithm, a slightly weaker conditional bound would still remain.
Why does anything have value? Ultimately because we can use them to get something we need or want. For the former, there is some sense in trying to reach an objective definition of value. The latter is inherently subjective. Some things are valuable to some people, because those people find those things valuable. That circular definition is the root of subjective value that cannot be reduced any further.
Bureaucracy has no direct value, but it may have indirect value, if it helps us achieve things that are more directly useful. Pure mathematics has had a lot of indirect value until now. It turns out that training people to do mathematics for the sake of doing mathematics, at any level from schoolkid to mathematics professor, teaches skills that transfer to other domains. If human thinking becomes obsolete, that indirect value may be gone, but the direct value will remain.
The results in pure mathematics are largely irrelevant, except as means of acquiring transferable skills. Sometimes they have practical implications, but such situations are exceptional. And there is not much correlation between the practical value of a result and its value as seen by mathematicians.
Helsinki just completed a 10 km light rail line that cost ~€400 million, including some major bridges. That's maybe €10k per current and future resident in the areas connected by it to the city center. Spending that much for an option to have good public transit only makes sense if most of those people are going to use the line.
Science, and research in general, should mostly be concerned with effective truth. The best objective-ish explanations so far, but something that can always be challenged with new evidence. The difference to objective/social concepts of truth is that you should always be willing to question the assumptions, beliefs, and reasoning ability you and people who agree with you have. If you discover something that goes against established beliefs, you should always start by assuming that you made a mistake somewhere. And if other people discover something like that, and it looks like they are seriously questioning the discovery, you should also take it seriously.
In Afghanistan, the issue was that the country does not really make sense. The central government is usually weak and illegitimate, and it doesn't have effective control over most of the country. The US went in with a small force and quickly defeated the central government. But because they lacked the manpower to control the country, they created a new central government, which was again too weak and illegitimate for the task. And then the situation just dragged on, until the US lost interest.
As for the Vietnam War, cities rarely changed hands in the war before the US started withdrawing its troops. The US was unwilling to invade North Vietnam, because it was afraid of an outcome similar to the Korean War, where it would have to face China and/or the USSR directly. And North Vietnam was too weak to invade the South.
The same should be true at every level of specialization. The more specific your needs are, the more likely one of the candidates is about as good as anyone in that specialty.
Europe started rapid remilitarization as a direct response to the war. The increase in military spending in European countries not participating in the war is comparable to worldwide investments in AI data centers and infrastructure.
Then there is the fact that a permanent member of the UN Security Council is waging an old-school war of conquest. If we assume that the post-WW2 world order is gone and wars of conquest are legitimate business again, the world will reassess other aspects of that order as well. Including the idea that people have inherent rights beyond those their government is willing to grant and capable of protecting.