This is bending over backwards so hard I think you had to invent new geometry and discover new dimensions.
Fan cars are old. Nothing about them is new or special or unique. Trying to build infrastructure dependent on such a feature would be the dumbest thing since monorails.
List of cars I know to go over 90MPH. Supposedly. 2000 Dodge Neon with 800 pounds of man in it and only 3 out of 4 cylinders functional. 2004 Toyota Camry with broken O2 sensors, and it loves it. Anything by VW. Pretty much anything parked in your local highschool.
Most modern cars can hit 90 MPH with ease, that's what modern aero and 145HP being "underpowered" means! Supposedly cars are electronically hard limited to 115mph and I have never been able to confirm that in practice
Of course, this implies that you might be able to own some of these features, with the actual crime being using them. Like that Beetle with a helicopter jet turbine.
Fan cars are pretty old. From that perspective, the driving upside down part is "less impressive" but screw it, still cool.
I want to see it on a track going all out, but something about what they decided to focus on tells me nobody gives a damn about this actually doing anything performance car related.
This "condition" has always been extremely sketchy, and just hard to study as well.
As someone who "has" it, it's fascinating. One line of my family all independently developed a warning when popping champagne corks because it causes us all to jump!
If I am on the right track with believing I have a presentation of this condition, it could be related to some of our genetic lineages being hilariously bad in various ways, specifically for this, a type of neurodivergence that leads to strong presentations of ADHD, Autism, and other Anxiety disorders and similar. They usually have "sensory" issues.
I have a strong sensitivity to loud noises and sudden stimuli.
For the "They do things you shout at them" part, that too is fascinating. My specific gene pool has a high probability of "Demand avoidance", but for me there is a strange twist that I have a demand compulsion for people I am close enough with.
I've always thought I "liked to help people" and liked fixing things and caring for others. I have a trait where if a significant other or family member states a minor problem ("It's dark in here, I'm thirsty, it's a little cold"), I have a strong compulsion to fix that. A previous partner made a point of telling me that I don't have to fix her problems, but I found it physically difficult to not do so even when told directly that I didn't have to!
My brain is all sorts of broken and wired wrong, from genetics but also because our lineage basically is a cycle of dysfunctional families raising dysfunctional families. So how much of this is my brain rewiring itself for a hostile environment from a young age I cannot tell.
I'm not convinced there actually is a relation to logging other than there being a large portion of this specific set of family lines in that industry from being Acadian.
It's really hard to separate any of our issues from any of our other issues. Is it related to our very high percentage of autoimmune disorders? Our generally high intelligence and adaptability? Who knows.
I don't think we should try to split accountability at all here.
The ad platform that allows malware to be served to anyone is at fault.
As is the individual website that has selected to allow a 3rd party's code to be delivered to your machine without controlling that risk, often for very silly reasons.
Oof, right at the top it points to how people were apparently convinced of this situation from a single crappy example
You work in data science and you over indexed on a singular example to explain away an entire set of phenomenon??
My experience with more.... directly testable NN models that are small seems to be that they absolutely can be far too interconnected to make strong explanations. The structure and pathways they fall into don't have to make inherent sense to anything or anyone.
Is it possible that almost all nodes in a network being explainable with a singular feature that makes obvious sense to us is evidence of poor efficiency? The network hasn't been made to collapse learning or discover more effective organizational structures than what might make sense to us.
I don't think I have the fundamentals here to contribute though.
Working with some neural networks, you can somewhat feel the structure when you read between the lines of the output. You can sometimes feel that a threshold somewhere tripped and set the signals down an alternative path.
But confirmation bias and all that.
I hope interpretability has a breakthrough. It would make my job far easier and cooler.
I had figured the "Recursive generalization" was like, understood? Maybe I'm just too pop-sci brained. "Split categories into more on this new boundary" and "Combine two separate categories that aren't different enough" I thought were always the core verbs for working with bulk data. My mental model has never been formalized though, so is invalid.
I guess I never got a good sense of the examples they had which made it look like what was "obvious" or made sense to me wasn't always a good explanation in general. Neat.
It's interesting that meaningful examples both ways care about shapes so much, as your eyes encode edge finding and shapes really early in the signal path. It's why some optical illusions are so effective. It's awesome.
The whole "how do you test a neuron: Find a group of nouns and find where you can split them" is just how I find I think learning works. You build your mental model, and find the split points where it diverges from reality and you are wrong, and you refine. I feel like this is how I learned Random Forests conceptually, though surely that's not a valid comparison?
Apologies, not sure I have actually contributed anything here! I bet I misunderstood the entire article somehow. Specifically, I do not have the fundamentals of looking into visual models that he is really working from. But I always figured that when Data Scientists talk of "Features" in larger models, they explicitly meant these far more abstract representational "features". Doesn't the premise of model hijacking and abbliteration sort of presuppose that you have abstract "Reject" or "Safety" neuron paths? I've always understood this stuff as that neuron path, without weighting too much on individual neurons.
>I used to think of neurons as human-shaped nouns: one is a car, another is a curve.
I didn't think anyone was ever supposed to think this for human neurons! The way humans identify something as even a "Chair" is insanely abstract and cultural! Very specifically, the "building sub branches to disambiguate" is just, like, how you would expect a hierarchical human brain to self organize, right?
Some of this is why I've been so displeased with the interpretability research from Anthropic, where it felt like they just pointed new NNs at the disparate neurons of another NN and post hoc named them. I felt the paths were of importance at least but had no idea how to convey, interpret, or model that. Learn you a good math kids! Not having any formalisms hurts.
There are good theoretical reasons to think superposition exists in neural networks, particularly in the residual stream rather than specifically in MLP neuron activations. It wasn't just a single example that got people thinking about superposition.
>OK, but you know Biden did more or less the same right?
No he did not.
Biden had some leftover documents that he turned over when told to. That is normal. It is quite literally part of secrets management to expect some stuff ends up uncontrolled and have processes and plans to recover it.
Trump had a secret stash of literal nuclear secrets in his meeting place with pretty much anyone willing to pay him, and denied having them, refused to turn them over, and denied everything again when he was finally raided and they found all that shit. People like you still insist somehow that this wasn't absurd on it's face.
Trump is the most sued president because Trump historically gets sued by everyone he talks to because, as he has literally told you to your face, that's his strategy, abuse everything and force it to go through the courts, because then you can bend those rules and slow down problems.
It's literally in the first sentence of the article you posted. "attorneys for Biden discovered classified documents" and contacted the authorities. How does that mean "got caught"? Your own attorneys can't "catch" you.
" Biden's personal attorney said on January 21 that the Justice Department discovered six items containing classification markings during a consensual search of his home the previous day, some of which dated to his tenure in the Senate; investigators also seized some of Biden's handwritten notes from his vice presidency."
"On November 22, 2022, Biden's attorneys discovered the first set of classified documents in a locked closet...they reported them that day".
If you're in possession of a calendar, you will notice that November 22, 2022 is actually before January 2023. You might also note the "consensual search" in your own quote.
So a normal person might conclude that the Justice Department called Biden after he reported on himself, and said "We see you said you had some documents, so maybe you forgot some more? Can we come search with you?" And Biden maybe said "Good idea."
reply