IP Reputation is only as meaningful as the duration of ownership. If it's the same owner for years, then reputation is meaningful, and that should count; if it changes hands every 6 hours being assigned to VPS clients or whatnot, then make the reputation stick to the /24 owner, and so on, with varying degrees of scope and duration, so that the responsible party - the shady companies renting their IPs to bad people - actually have their reputations stick. Then block the /24 or larger subnets, or aggressively block all ranges owned by the company, isolating them and their clients, good and bad.
That sort of pressure can work. But then you risk brigading and activist fueled social media mobs and that's definitely no way to run the internet.
What's the purpose of blocking them, anyway? Is it to make you feel good? To clean up logs? To reduce spam? With the residential proxy industry - which, I note, is directly boosted by such blocking practices and funnels money into organized crime - IPs don't mean a whole lot to those who can pay.
100% agree with your point regarding long term ownership allowing for meaningful reputation.
I don't necessarily think that's 'no way to run the internet' or even 'no way to run anything', in that people can choose to whom they listen in regards to blocking, protesting, boycotting.
As long as none of the different groups of opinions are forced on anyone else, then pick and choose those you apply and those you ignore.
With my lists of blocking, I classify them, personally, into different tiers such as Basic, Recommended, Aggressive, and Paranoid when I apply the rules to other people's (family) setups - I'm the only one that uses Paranoid.
Large DDoS botnets will have hundreds of thousands of return-path-capable IP addresses. Your temporary blocks will have to be very sensitive (i.e. trigger on a relatively small number of requests within the time window) for an application-level DDoS to be usefully mitigated.
Preventing 1% of sunlight from hitting Earth is more than enough to offset climate change heating. It's not enough to make agriculture or photovoltaics uneconomical. In many regions, it might make agriculture more viable on the net, not less - by reducing climate risks.
Only becomes viable if you have things like Starship online and fully operational, with launch rates at the level of Falcon 9 today. At the minimum.
Still a more viable option than bringing greenhouse gas emissions into the negatives globally, by the way. But that's a low bar. Nuking the ocean floor is probably a better call.
That's the sum of climate change. "GDP growth of 3% instead of 5%."
Severe enough to be noticeable, but not severe enough to warrant radical climate action. Not an extinction threat. A "slow trickle of economic damage, some amount of otherwise preventable death and suffering, diffused across the entire world, applied unevenly, and spread thin across many decades" threat.
And stopping the GHG emissions demands radical, coordinated global action. Major emitters would have to pay local costs now - for the sake of global benefits many decades down the line. And those emitters are not the countries that face the worst climate risks. Global superpowers can tolerate climate change - it's countries that already struggle as it is, that don't have the resources to adapt or mitigate damage, that can face a significant uptick in death and suffering rather than damage in the realm of economics.
That makes climate action a very hard sell for the politicians. Thus the tepid response.
By now, I'm convinced that the only viable approaches to climate change lie in the realm of geoengineering. Which does not require multilateral coordinated action against a "tragedy of commons" scenario, and is cheaper than forcing local GHG emissions into negatives.
Even non-permanent geoengineering solutions offset impacts here and now - thus buying time for fossil fuel energy to succumb to the economic advantage of renewables. And geoengineering measures can be enacted unilaterally by many powers - as long as the political will is there to absorb a few strongly worded condemnation letters.
And then when the GDP finally collapses, there will have been nothing that could be done about it for the last 50 years and they'll ask wtf we were doing in 2040, why we didn't stop it then.
That's the nasty thing. It doesn't "finally collapse".
The world just eats the climate costs and keeps going.
There's no global catastrophe. No single moment when the magnitude of your folly is revealed to you a blinding flash. Just a slow trickle of "2% worse". A loss of what could have been.
Whether there will be a global catastrophe or not, it is unknown yet and unpredictable.
There are many mechanisms of positive feedback that can accelerate global warming instead of just reaching an equilibrium at a higher average temperature than now.
If some of those mechanisms of positive feedback would be triggered, a global catastrophe would be possible, due to the excessive speed of the climate change, which does not give enough time for the biosphere to adapt to it.
Instead of hoping that we will be lucky, it would have been much better to avoid such risks and prevent further increases in CO2 concentration and average temperature.
I am old enough to have seen a dramatic change in climate from the time when I was a child, when the seasons were still exactly as they had been described for centuries and millennia at that location in Europe, to the present time, when winters are no colder than autumns were before and I have never used again my winter clothes and boots for about 15 years.
I find such a radical change during my lifetime quite scary and I see no evidence for claims that "the world will keep going". The truth is that nobody knows whether this will be true and hoping that this will happen without doing anything to guarantee such an outcome is reckless.
There is not enough "positive feedback" going around in the system to do anything of the sort.
Plenty of feedback mechanisms were proposed, investigated, and found lacking. It's a "makes climate change 10% worse than it would otherwise have been" kind of thing, not a "makes climate change 1000% worse than it would otherwise have been" kind of thing.
The world got hit with WW2 and moved on. It takes a lot to destroy "the fundamentals of civilization" on a global level. Climate change is woefully insufficient.
The problem is that we will not move on from WW3, and famine, water depletion, resource exhaustion etc. are all existential problems for individual countries that will cause conflicts between previously peaceful nations. At some point the nations in conflict will have alliances and nuclear weapons, and people will use them when the choice is between that or starving to death by the millions. I would be somewhat more optimistic about humanity's ability to weather worsening circumstances if we didn't develop the human extinction button in all of our grand technological wisdom.
Climate change is not the great equalizer people want it to be.
Nuclear superpowers are among the least likely countries to actually collapse from climate damage.
US isn't Syria, and it's Syria that's at risk.
First world countries like France can absorb a +30% spike to food prices. Countries where the same food price spike would come with a major death toll don't have the tools to kick off WW3.
Pakistan and North Korea have nuclear weapons. What makes you think that other countries won't develop them when push comes to shove? Right now the status quo is such that smaller countries find violating international sanctions on nuclear weapon development to be disadvantageous (no immediate benefit from having them; expensive to have them; economic punishment for having them). The calculus on the status quo changes considerably when famine or other ecological disasters are threatening to wipe out half your population. Is the US going to invade all potential nuclear weapons developers like it did with Iran? Do you have complete confidence that will always work and never escalate towards anything larger?
You know it. The country that was probably the closest to "developing nuclear weapons" is currently in the process of being bombed to shit for it.
Not the only reason, no. But their nuclear ambitions certainly did contribute to their current predicament.
Nuclear weapons take time, expertize, resources and sustained political will to develop and deploy. If you have all of that, you might be able to put all of that towards climate adaptation instead. Or: go for the nukes and bet hard that you aren't going to get bombed to shit for it. Worked out for NK, but doesn't seem to have worked for Iran.
I think the level of tolerance US has for marginally stable autocracies with nuclear weapons has receded permanently.
Not to mention that merely having a nuclear bomb doesn't automatically allow one to cause WW3. Nuclear weapons were used in a world war already, and that didn't even destroy a single country - let alone the world. It takes a lot of nukes, and a lot of delivery mechanisms, to actually move the needle on the matter of human civilization existing.
> Is the US going to invade all potential nuclear weapons developers like it did with Iran?
Invade? No. Bomb? Probably. Same if for India if e.g. Sri Lanka decides it wants nuclear weapons.
Global warming will unfortunately disproportionately hit poor, equatorial countries. (Also, starving countries can’t afford a nuclear programme. There is no breakout risk in Sudan.)
WW2 hit the higher layers of abstraction. You could still grow food plants in those countries during and after the war, but the governance structures were bombed. Climate change is the opposite. If you can't grow food, you're fucked.
Although I live in the EU, I have no trust in its ability to regulate my media usage or platform providers at all.
The EU just managed to postpone chat control for a bit, and my own country has found a renewed passion for punishing expression crimes (so-called "Äußerungsdelikte") through various legal and pre-legal means.
Social, legal or technical centralization is not a solution to any issue related to public discourse, and Euro-nationalism is not a wise concept. It will simply make us another economic bloc, just with an older population than the others.
Contrary to the current zeitgeist in the EU, power should be dispersed as much as possible. We should embrace global open-source initiatives and work towards a European Union that open-source projects (and tech companies!) want to organize under because of our superior regulatory frameworks, not subsidies, legal pressure, promised government service demand or political initiatives.
We already have a lot of failed political initiatives, so why not try the organic, good governance approach for once?
Instead, we just create more bureaucracy and red tape. This absurd CRA nugget is a good example for our european style tech regulation for open source: https://cra.orcwg.org/faq/stewards/
SpaceX is the undisputed king of launch cadence. Falcon 9 just flies every other day nowadays.
If anyone can take "we need 14 launches per mission" and make it work, it's SpaceX.
Boil off isn't somehow unsolvable. We know cryogenics can work in space, and SpaceX's approach is actually less aggressive than Blue Origin's requirement of zero boil off on LH2.
The only reason NOT to cancel SLS outright is "we can't get anything better". "Sure, it's pretty dumb, but it can be built, and good luck getting anything better built."
Starship is important because the closer Starship gets to coming online the more obviously wrong that line of thought is.
As is, Starship, with its first stage being online and reusable already says "we could have done something like SLS much cheaper if we were smart about it". When the second stage comes fully online, the argument for SLS will diminish further.
This not actually a good reason. First of all, of course you can't get anything better if you never ask for anything better or consider alternatives or put any money into anything better.
If you never invested in anything else, then the M4 Sherman tank would still be the best tank. And then you could say 'we can't get anything better' and continue to use it while refusing to ever even put 1$ into developing anything else.
And actually there are plenty of ways, even without Starship to do these things differently.
I remember when SLS fans in 2016 told me that SLS is real and Falcon Heavy is fake. Even when Falcon Heavy and now New Gleen can do most things SLS can.
One NASA Administer was almost fired for exploring if Orion could launch on Falcon Heavy.
The thing is NASA is not looking for alternatives, and it doesn't matter if Starship is fully proven and operational. People will still say its not the same, because Starship will need refuel and isn't direct.
You can always find a reason to justify one solution if you only ever consider that solution and refuse to even look at any other possibility.
It might work, I considered running a test like this. But it does demand certain things.
The subnetwork has to be either crafted as "gradient resistant" or remain frozen. Not all discovered or handcrafted circuits would survive gradient pressure as is. Especially the kind of gradients that fly in early pre-training.
It has to be able to interface with native representations that would form in a real LLM during pre-training, which is not trivial. This should happen early enough in pre-training. Gradients must start routing through our subnetwork. We can trust "rich get richer" dynamics to take over from there, but for that, we need the full network to discover the subnetwork and start using it.
And finally, it has to start being used for what we want it to be used for. It's possible that an "addition primitive" structure would be subsumed for something else, if you put it into the training run early enough, when LLM's native circuitry is nonexistent.
Overall, for an early test, I'd spray 200 frozen copies of the same subnetwork into an LLM across different layers and watch the dynamics as it goes through pre-training. Roll extra synthetic addition problems into the pre-training data to help discovery along. Less of a principled solution and more of an engineering solution.
+1 I’ve always had the feeling that training from randomly initialized weights without seeding some substructure is unnecessarily slowing LLM training.
Similarly I’m always surprised that we don’t start by training a small set of layers, stack them and then continue.
Better-than-random initialization is underexplored, but there are some works in that direction.
One of the main issues is: we don't know how to generate useful computational structure for LLMs - or how to transfer existing structure neatly across architectural variations.
What you describe sounds more like a "progressive growing" approach, which isn't the same, but draws from some similar ideas.
In terms of sub structure - in the old days of Core Wars randomly scattering bits of code that did things could pay off. I’m imagining similar things for LLMs - just set 10% of weights as specific known structures and watch to see which are retained / utilized by models and which get treated like random init
No. You cannot. It's the wrong tool for the problem.
That little "add" of yours has the overhead of: having an LLM emit it as a tool call, having to pause the LLM inference while waiting for it to resolve, then having to encode the result as a token to feed it back.
At the same time, a "transformer-native" addition circuit? Can be executed within a single forward pass at a trivial cost, generate transformer-native representations, operate both in prefill and in autoregressive generation, and more. It's cheaper.
reply