The big news this week was Apple Intelligence being integrated deeply into all their products. Beyond that, we had a modestly better than expected debate over the new version of SB 1047, and the usual tons of stuff in the background. I got to pay down some writing debt.
> Or they embed malware in a Stable Diffusion extension pack and tell victims ‘You have committed one of our sins: Art Theft.’
That is not at all what happened. The dev (AppleBotzz) is the one who embedded the malware (and then modified the package twice to make it harder to detect). Then he tried to false-flag it, making up a nonexistent anti-AI hacker group and pretending they were responsible. This can be confirmed easily by looking at the repo history. The malicious code was already in the very first version of the code.
This is a very ordinary "malware creator creates and distributes malware" story, it has nothing to do with AI haters.
I suppose AppleBotzz's true motivations are known only to himself, so for all I know he *could* have been acting out of anti-AI sentiment, but the malware is designed to steal credit card info and login credentials, so a straightforward criminal financial motive seems much more likely. He is definitely lying about the repo having been hacked, though: it was malware when it was created, and AppleBotzz made several edits to it that were designed to make the malware harder to detect.
> Achieve Artificial Gerbil Intelligence achieved internally at DeepMind, via a deep reinforcement learning to train a virtual agent to imitate freely moving rats with a neurochemically realistic model of their brains. Insights are then gained from looking at the model. Gulp?
My impression (I haven't looked at the paywalled paper itself though!) is that they trained a virtual agent to imitate freely moving rats and then the internal activations of the trained neural network were a much better predictor for real rat neural activity than the movements themselves.
I think this might be a good example for the idea that neural networks learn the generators of the data they're trained on and not just the stats.
> To have 1,000 new companies of 100 brilliant people, that means (if you consider failures and early stages) 3% or more of them have to become founders.
This would also drastically drive down the returns to being a founder. There's only so many $5M penthouses available in Silicon Valley. If becoming a founder no longer means you have at least a 1% chance of getting that penthouse, a lot less people will want to do it.
I appreciate the introduction of expected costs of insurance into the discussion, even though the worst possible outcomes are uninsurable.
> 'better dead than red'
Coincidentally, 'red' works for Chinese/Communist victory or for making a loss ('in the red'). Perhaps everyone else already noticed.
That reminds me of one big thing I don't know: does the Chinese government try to steer the most talented young coders into machine learning, hacking/espionage (whatever the target), or something else? Assuming it tries to steer them at all.
Looking forward to Zvi's take on Aschenbrenner's positions.
"As for AI, I could ask if you (and your own AI) can’t tell it’s AI, what’s the problem?"
I figured the problem with getting unlabeled calls and messages from an AI is that it lowers the cost for someone to waste your time. However cheap it might be to hire a human to make unsolicited or phishing calls, it's even cheaper to get an AI to do it. Lowering that barrier is a problem, so there is value to AIs being identifiable.
> We reformed the entire healthcare system with a bill where the Speaker of the House said ‘we have to pass the bill to find out what is in it.’
Per Snopes[1], this quotation is often used in bad faith. Pelosi meant, roughly, "once voters sees what is in the bill as opposed to being confused by misleading descriptions, they will like it." She was right.
> Or they embed malware in a Stable Diffusion extension pack and tell victims ‘You have committed one of our sins: Art Theft.’
That is not at all what happened. The dev (AppleBotzz) is the one who embedded the malware (and then modified the package twice to make it harder to detect). Then he tried to false-flag it, making up a nonexistent anti-AI hacker group and pretending they were responsible. This can be confirmed easily by looking at the repo history. The malicious code was already in the very first version of the code.
This is a very ordinary "malware creator creates and distributes malware" story, it has nothing to do with AI haters.
Huh. Had no sign of that. Can anyone confirm (or can we have a source)?
There's detailed discussion of the incident here: https://www.reddit.com/r/comfyui/comments/1dbls5n/psa_if_youve_used_the_comfyui_llmvision_node_from/
I suppose AppleBotzz's true motivations are known only to himself, so for all I know he *could* have been acting out of anti-AI sentiment, but the malware is designed to steal credit card info and login credentials, so a straightforward criminal financial motive seems much more likely. He is definitely lying about the repo having been hacked, though: it was malware when it was created, and AppleBotzz made several edits to it that were designed to make the malware harder to detect.
Multi Voiced AI Narration of this post:
https://askwhocastsai.substack.com/p/ai-68-remarkably-reasonable-reactions
FYI, “Divida Eden” is a typo; her name is Divia.
Yeah, I actually know her. Normally names are mental typos, I have no idea how THAT one happened.
Zvi, the post says "Need Nanda" instead of Neel 😂
> Achieve Artificial Gerbil Intelligence achieved internally at DeepMind, via a deep reinforcement learning to train a virtual agent to imitate freely moving rats with a neurochemically realistic model of their brains. Insights are then gained from looking at the model. Gulp?
My impression (I haven't looked at the paywalled paper itself though!) is that they trained a virtual agent to imitate freely moving rats and then the internal activations of the trained neural network were a much better predictor for real rat neural activity than the movements themselves.
I think this might be a good example for the idea that neural networks learn the generators of the data they're trained on and not just the stats.
(Someone please double-check the paper, though.)
Volcano power plants are already in the works - https://www.newsweek.com/drilling-volcano-energy-renewable-magma-1860257
Those Kelsey Piper excepts are *chefs kiss.*
> To have 1,000 new companies of 100 brilliant people, that means (if you consider failures and early stages) 3% or more of them have to become founders.
This would also drastically drive down the returns to being a founder. There's only so many $5M penthouses available in Silicon Valley. If becoming a founder no longer means you have at least a 1% chance of getting that penthouse, a lot less people will want to do it.
I don't actually think most of them do it for the penthouse (in the broad sense).
I appreciate the introduction of expected costs of insurance into the discussion, even though the worst possible outcomes are uninsurable.
> 'better dead than red'
Coincidentally, 'red' works for Chinese/Communist victory or for making a loss ('in the red'). Perhaps everyone else already noticed.
That reminds me of one big thing I don't know: does the Chinese government try to steer the most talented young coders into machine learning, hacking/espionage (whatever the target), or something else? Assuming it tries to steer them at all.
Looking forward to Zvi's take on Aschenbrenner's positions.
"As for AI, I could ask if you (and your own AI) can’t tell it’s AI, what’s the problem?"
I figured the problem with getting unlabeled calls and messages from an AI is that it lowers the cost for someone to waste your time. However cheap it might be to hire a human to make unsolicited or phishing calls, it's even cheaper to get an AI to do it. Lowering that barrier is a problem, so there is value to AIs being identifiable.
Very informative, thank you.
> We reformed the entire healthcare system with a bill where the Speaker of the House said ‘we have to pass the bill to find out what is in it.’
Per Snopes[1], this quotation is often used in bad faith. Pelosi meant, roughly, "once voters sees what is in the bill as opposed to being confused by misleading descriptions, they will like it." She was right.
[1] https://www.snopes.com/fact-check/pelosi-healthcare-pass-the-bill-to-see-what-is-in-it/