Quick iteration is definitely the big thing. (The eye is fun because it's so "badly designed" - we're stuck in a local maxima that just happens to be "good enough" for us to not overcome the big glaring problems)

And yes, if all the inputs are corrupted, the output will likely be too. But 1) they won't all be, and as long as there's a good mix that will "teach" the network over time that the difference between a "corrupted cat" and an "uncorrupted cat" are irrelevant, because both will have most of the same labels associated with them. 2) these tools work by introducing corruption that humans aren't meant to notice, so if the output has the same kind of corruption it doesn't matter. It only matters to the extent the network "miscorrupts" the output in ways we do notice enough so that it becomes a cost drag on training to train it out.

But you can improve on that pretty much with feedback: Train a small network to recognize corruption, and then feed corrupted images back in as negative examples to teach it that those specific things are particularly bad.

Picking up and labelling small sample sets of types of corruption humans will notice is pretty much the worst case realistic effect these tools will end up having. But each such countermeasure will contribute to training sets that make further corruption progressively harder. Ultimately these tools are strictly limited because they can't introduce anything that makes the images uglier to humans, and so you "just" need to teach the models more about the limits of human vision, and in the long run that will benefit the models in any case.

As long as people aren't ready for it, then it doesn't solve the immediate problem that needs to be solved today.

You wouldn't want to. If you just feed it to the models, then if there are enough of these images to matter the model will learn to ignore the differences. You very specifically don't want to prevent the model from learning to overcome these things, exactly because if you do you're stuck with workarounds like that forever, but if you don't the model will just become more robust to noisy data like this.

Pet dogs also eat poo on occasion, also without any underlying problem, so I really don't think there's any reason to think that far less domesticated species where it is well established would just stop. I'm sure you can reduce it, especially if it has a nicer food source, but still, an animal with far less history of domestication seems like a recipe for amplification of all the potential issues you don't want to deal with.

That is a reason for arguing that people don't always make smart choices. It is however not an argument for claiming how people vote does not show what their preference is at the time of voting, which is what is relevant here.

It's perfectly fine to argue you think it's stupid of people to want to read about Musk, but the votes clearly show they do in fact want to.

If all they did was argue for jettisoning stone age thinking, I'd have been all for that. But instead the person I replied to engaged in just another variant of the same stone age thinking.

If the Palestinians were actually interested in stopping their oppression, they would stop trying to fight an insurgency against the Israelis. As it is, they are a security threat, and for good reason.

Which oppressed peoples have come well out of surrendering to a party that has refused to give concessions?

This is classic victim blaming. It's also demonstrably false: There have been many lullls in the fighting. And yet the Israeli oppression has never stopped. If anything, Israeli has continued the war crimes of settling occupied territories, and the crimes of Apartheid by tightening the control of the borders of the occupied territories and limiting the movement of the Palestinian position, as well as ramped up racist laws such as the nationality law, and this expansion of oppression has never once stopped when resistance has abated.

This notion that you can end oppression by appeasing your oppressor is not one that has a very successful history in general, and Israel has proven time and time again over a period of decades that it definitely does not work with Israel.

And irrespective of that you fall in the typical trap of thinking you can talk about Palestinians as a unified, single entity, rather than as a mass of people with different views where even if 99.9% were to suddenly decide they trusted Israel would treat them fairly if only they stopped fighting, that would not stop the remaining 0.1%.

Notice how you yourself de facto treats Hamas as a proxy for Palestinians as a whole:

I don’t see why Israel should give quarter to Hamas now, nor should they entertain the idea that Palestinians are being sincere in co-existing with Israelis.

Consider e.g. the IRA, which saw support diminish substantially (while Hamas' support is still high), and yet still continued an insurgency in a far less oppressive situation until the UK government sat down and actually listened to their concerns and gave concessions.

Israel has created a population where sufficient numbers of people feel they have nothing to live for. There is no realistic scenario here where the insurgency ever stops unless Israel commits total genocide or seeks a negotiated settlement including giving substantial concessions irrespective of whether or not they think they can trust any of the parties.

That is not a question of whether that is fair, or reasonable, or whether it's the smart think to do for Palestinians to continue.

It is what will happen when you create a situation like this.

So you talk about what Palestinians are "actually interested in", but the Palestinian people as a whole have zero power to end this because it'd require the total agreement of each and every one of five million individuals. Israel on the other hand has the power to end this, because on their end it only requires the agreement of the state to dial back the oppression enough that support for groups like Hamas loses support, and then negotiate an end to it.

It would be so much easier if Israel just considered them as the enemy, and throw them out of the territory of Israel as they wished. It’s only right for a bunch of sore losers. Let them resist from outside the territory of Israel proper, and seek help and refuge from their Arab “allies”.

Not even Israel considers the occupied territories theirs. They are not the territory of Israel even under Israeli law. As it is what you're suggesting here would be a severe violation of international law, a crime against humanity, a violation of a number of UN decrees, and would violate Israeli law as well, as Israel's actions are only accepted by their own Supreme Court on the basis that Israel's own government have consistently insisted it's done under a doctrine of belligerent occupation: In other words, they do consider them the enemy and despite their many war crimes, not even Israel is prepared to commit the level of war crimes you suggest.

It is fairly fascinating yet also shocking how many people here argue for a maximalist position so extreme that even the far-right Israeli government rejects it as too extreme, as have every Israeli government since 1967.

The Palestinians are probably not that bad of a people, but it doesn’t help that they keep making armed struggles for a lost cause when they can just make peace on their loss.

Telling people to just lie down and let their oppressor keep oppressing has historically never worked. Israel will never experience peace without coming to terms with that, because every new generation growing up in those conditions will learn from a young age to hate.

Nice "whatabout", but the bully here is the party that engaged in an illegal occupation, the crime of apartheid, and extensive war crimes (annexation through settlement of occupied territory) in the first place. That you try to redefine away the fact that Israel created this situation in the first place borders on apartheid-apologism. It's exactly the same tactic used by supporters of South African apartheid to dismiss the situation in South Africa whenever the ANC carried out a violent operation, and it was apologism for oppression then, and it is apologism for oppression now.

Not everything is on Github yet, and much of it is messy and with dependencies on my environment (in the process of cleaning that up), but if you click through the "cross-posted to:" link there are a bunch of links to repo's of what I have pushed so far.

It's "psychoactive" in the same way that caffeine is. That is, it's a stimulant. Using that term only serves the purpose of making it sound scarier. And it's far less addictive on its own than when smoked. It's not harmless, but it's also nowhere near as big a problem in itself as specific product categories and delivery methods, and no worse than any number of other things we're perfectly fine with people using.

Latency limits datacenter placement a lot, but for batch jobs like AI training it's certainly an option.