I'm not even sure where to start, but I have to ask- are there any groups / projects that are trying to make and train an OSS model federated across members' machines?
Someday I'm going to write a paper that achieves SOTA results with a nigh-incomprehensible mishmash of diverse techniques and title it "All You Need Considered Harmful".
The most promising work along these lines centers around augmenting LLMs with an external data store ("retrieval-augmented LLM"s). I think this started with Facebook's KNN-LLM ( https://arxiv.org/pdf/1911.00172.pdf ). Legal conflict may force the industry to move towards vector DBs as the predominant method by which facts are "stored" rather than model parameters ( https://arxiv.org/pdf/2308.04430.pdf ) , with the happy side effect of update-ability over time.
These are good examples of mass casualty events that didn't motivate effective responses. For both climate change and guns, there are people that want to keep the status quo as it is. I believe that popular arguments against gun control will be able to stay effective in ways that arguments against climate action will fail.
First, one of the most popular arguments against action addressing climate change has been that climate change is simply fake. A mass-casualty event would make that argument less effective.
The gun control debate in the United States is linked with people's perceptions of personal agency. Often, a mass casualty event perpetrated with a gun is ended by a gun being used on the perpetrator (sometimes, it's the perpetrator killing themselves, and sometimes it's police killing them). Many people imagine it would be possible to end those events as a bystander with a gun. But I think that it would be much harder to get people to fantasize about being a hero that stops a mass casualty event resulting from extreme heat. Such an event starts and ends when Mother Nature says it does, and it seems like it should be pretty clear to people that the only way personal agency is relevant is in the act of preparing the environment.
News media like to have stories they can focus popular attention on, and no one car accident (unless it kills someone famous) will matter to enough people to warrant broader coverage than one's local paper. Despite this lack of media visibility, there actually is some movement at the city level to change the design of roads to reduce the lethality of accidents. This isn't very publicly visible yet, I don't think, and there's certainly very little in the way of public desire to drive less lethal vehicles.
I haven't read the book, but if the event the book describes kills hundreds of people or more, it seems like a plausible seed that news coverage can accrete around.
I'm curious about whether anyone has studied using the harmonic mean (or at least, the reciprocal of the sum of reciprocals) as a way of aggregating utilities. I haven't had time to research this, but the thought repeatedly occurs to me as I have kids that are gifted in a school district that is especially concerned with the gap in achievement between the highest and lowest performers. I can't shake the feeling that what they really should want is a measure that prioritizes helping the lowest performers but does not consider the performance of the highest performers to have literally negative value, and the harmonic mean fits that bill.
A group being "above the rules" is, I think, a statement about what the rules are and how they are enforced. It doesn't really hinge on whether any members of that group have, up to this point, violated the rules.
As others have said, if they're limiting the number of approvals you can make on your ballot, then that's not how approval voting is supposed to be administered.
I have long liked and advocated for approval voting (or cardinal methods more generally), or barring that, basically anything but FPTP.
But I'm disillusioned. Many non-plurality methods (including approval) can effectively solve the spoiler problem (i.e. satisfy IIA)... but I no longer think that spoilers are the absolute most important problem to solve.
The most important problem is single-seat districts creating anti-majoritarian, disproportional legislatures. Because of how people geographically sort themselves among the likeminded, even independent districting commissions will not be able to effectively prevent anti-majoritarian legislatures.
For this reason, I believe that we would greatly benefit from explicitly proportional representation. This can be accomplished through cardinal or ordinal means - I no longer care about that dimension. That doesn't mean that we wouldn't enjoy some incremental benefit from approval or really any non-FPTP system. We would. Seattle should go ahead and do its thing. But the fight for better forms of democracy can't stop with any system that retains single-seat districts.
I have used genetic algorithms to search for non-transitive die triplets among the standard set of platonic dice, and exhaustively searched d6 and d8 numberings. If one insists that the advantage of A over B be the same as the advantage of B over C and C over A, then 61% is the best (farthest from 50%) that can be achieved for d6 and d8, and the best that I have been able to achieve for any larger dice.
Maybe the article could have included that if an automotive engineer had written it. But I think the article is perfectly fine for having included the steps that it did address.
Think about it this way. If you wrote an article about preventing the Toyota bug that resulted in uncontrolled acceleration, unless you were a former Toyota employee there's simply no way you can say "Toyota needs to change these specific lines of code" and you wouldn't even try. You might say "Toyota needs to adopt such-and-such code review practices" or "NHTSA needs to regulate the computer-controlled components of cars like this". Those are early, indirect steps that should force (or incentivize) the actual detailed improvements that you'd like to see happen.
The article talks about solutions at three different levels, though perhaps more indirectly than you were looking for.
The end of the fourth paragraph and beginning of the fifth paragraph talk imply a solution: use crash test dummies that accurately simulate women's anatomy. (This is made a little more explicit in the ninth paragraph, as it mentions "requiring updated and more equitable dummy implementation tested in every seat".)
The seventh paragraph mentions that insufficient regulation of the auto industry has allowed them to build very large cars. It is thus implied that regulating the auto industry to produce smaller cars would help.
The eight paragraph talks about how the exclusion of women from automotive design decisions has lead them to ignore women as consumers (and crash victims). The implied solution is for auto companies to hire more women.
It's not just a problem that he's discussed before- it is the central motivation behind his approach of "person-on-the-street interviews".
His view (supported by his grad research) was that the default approach of presenting information allows viewers to gloss over how the video's information differs from their pre-existing mental models. After the video is done, they can go back to their pre-existing models without actually updating with info from the video, because as far as they're concerned, there wasn't conflict between the pre-existing model and the video's info.
His hope in showing interviews with people off the street is to get people to think about how they might answer the interview question. Get people to remember their pre-existing model. Then hit them with the contrast between their pre-existing model and the truth, so the viewer hopefully makes an attempt to incorporate the truth.
Now, this doesn't say anything about the information density of his videos. I don't think they're very information dense. This reply is just about his attempt to increase the viewer's retention of the information that is presented.