Depends on the definition. It's become more frequent for predators to download images of real children from their parents' social media, then edit them with diffusion models to be nudes. This should still be illegal, and it is in the US.
> In People v. Gerber (2011, 196 Cal.App.4th 368), the California Court of Appeal overturned a conviction against a father who digitally pasted his 13-year-old daughter's face onto adult pornographic images. The court ruled that this did not violate California Penal Code §311.11, which criminalizes possession of child pornography involving a real minor "personally engaging" in sexual conduct. Since the images were digitally manipulated and did not depict the actual child involved in any sexual activity, the court held that they did not meet the legal definition of child pornography. The decision aligned with the U.S. Supreme Court's ruling in Ashcroft v. Free Speech Coalition (2002), reaffirming that purely virtual or fictional content—without the involvement of real minors—is protected by the First Amendment.
Most of the laws prohibiting deep fakes don't ban deep fakes on their own. They require specific other behavior, like using deep fakes to harass someone.
This is true, although it seems that the common denominator is that transmission will suffice, regardless of whether there is an intent to harass, blackmail, defame, etc.
Since almost anything involving the Internet can be held to be transmission, and indeed, interstate commerce, it seems that doing anything other than literally generating it locally on your own hardware and viewing it privately within the confines of your own home can be a serious violation of the modal state law on this issue.
Where laws really do seem vague--to my untrained eye--is on what constitutes sufficiently "sexually explicit" imagery. The traditional child pornography laws are pretty blunt on this, and mean basically anything that could conceivably arouse, it seems.
Here in Georgia, for example, the controlling legislation on non-child deepfakes seems to be SB78 (transmission of deepfake imagery with intent to harass, or posting publicly in general), which says[1]: "transmitting or posting nude or sexually explicit photographs or videos for purposes of harassing the depicted person; to provide for an increased penalty for the electronic transmission or posting to a web page that is accessible to the general public..."
But I suspect a lot of cases under this would turn on what exactly "sexually explicit" means.
MB has by far the worst leadership out of all German car brands
The new Porsche CEO definitely knows what he is doing, VW's core brand is also evolving well with the ID Polo, and even BMW has gotten something right with the Alpina acquisition and concept
Källenius on the other hand is responsible for the gigantic commercial flop that is the EQ series. He directed that before becoming CEO and then had to shut it down as CEO because people didn't see the cars as prestigious. And they are right. They look and feel like new Chinese brands. And don't even get me started on the new AMG GT. That is both a hideous and soulless car.
Exactly this, since the diesel scandal it is also legitimate to say that Mercedes and VW management is a criminal organisation. They have ruined both brands, taken insane money from German taxpayers and now they are leaving. It would have been nice if the government stopped supporting them earlier and rather helped ev startups flourish more.
Porsche's financials are trash because of their cancelled EV projects. After 12 years of half-assing electrification, it looks like BMW is really hitting it out of the park with the new i3/ix5/ix5.
The new Porsche CEO took over in January, the Mission S has been in development since ~ March and that is the first interesting new Porsche model in ages
Check the new cars, like the ID Polo, ID3 Neo and ID Cross. They've done a 180 on interior and fixed the exterior design language. It's one of the best interiors I've seen in a long time in entry level to midrange cars, and the way their CEO talked about it was refreshing.
This is the general mindset in Germany, atleast amongst people who follow new car releases. The ID Polo in particular will be a smash hit, I'm sure.
They say the difference between an optimist and a pessimist is that a pessimist says "It can't possibly get worse than this!", and an optimist says "sure it can!"
It's threatening the construction boom, but it's still mostly American companies that are benefiting from an AI buildout, no matter where it happens globally (with the exception of China).
Google announced yesterday that they are pouring an additional 15 billion USD into Malaysia for compute. ByteDance is also building there, through DayOne. If you think about it, there aren't many reasons against constructing AI inference capacity outside the US. You tend to have less local resistance and there are plenty of countries with sufficient grid robustness and internet backbones. Malaysia already has many datacenters so this isn't novel.
LLMs typically don't have high latency sensitivity since reasoning models came around, so you aren't forced to build them near users, and East Asia is seeing lots of adoption anyways.
So all in all, I think US legislators are overestimating their leverage here
> You tend to have less local resistance and there are plenty of countries with sufficient grid robustness and internet backbones.
What explains the reduced local resistance? Do Malaysian residents simply have less political power, or do Malaysian datacenters somehow solve the technical challenges that beget resistance in home countries?
Because Chinese models heavily rely on distillation and regularly behave strangely in public benchmarks (where they score much better than in private evaluations), even if many world class researchers are Chinese. It's difficult to quantify to what extend labs like Moonshot would be competitive if they could not distill from OpenAI, now that Anthropic mostly blocks it
You are fundamentally misinterpreting this paper. If this should hold up in review, it contains multiple capabilities that no human and no technology so far (especially not ROME) can do.
Experiments 1, 2, 3, 5 and 6. All of these are technical breakthroughs
Weight space programming is a holy grail of RSI research, and what you are talking about is very different from that, while the paper's proposals are exactly that.
It's similar in spirit to the Navier Stokes solution in that this level of LLM can intuitively navigate extremely large search spaces, rather than inefficiently bruteforce something
> Experiments 1, 2, 3, 5 and 6. All of these are technical breakthroughs
I haven't wrapped my head around 5 and 6 so maybe these are a bigger deal, but 1,2 and 3 are trivial. And the author's writing about 7 and 8 (which show that 7 is just directional ablation and nothing more), and the lack of prioritization (why put 7 and 8 in the paper at all if 5 and 6 are the big deal?) really doesn't inspire confidence in the paper as a whole.
> In Experiment 5, we ask Astra for networks that play three MinAtar games, and in Experiment 6, for a network that plays Go.
5 and 6 aren't surprising, merely impressive, a direct extrapolation of human abilityies. I've written transformer weights by hand to implement a simple shortest-path graph search. It takes hours, so an LLM that's >100x faster than me obviously can build super complex networks. It's just like creating any intricate system by hand such as writing machine code or designing computational complexity widgets.
On the other hand I'm amazed by how few thinking tokens Astra needed to do things like analyse networks in 1, 2, 3. That's real depth of reasoning: experiments have shown that Astra can perform a huge number of arithmetic calculations per single output token, double Fable or any other. Its brain is wired for mathematical reasoning
[Edit: looking closer at the full prompts and reasoning for experiments 1-3, I'm a bit less impressed seeing the shortcuts there. But looking at a large matrices of numbers and picking out patterns is much like looking at thousands of tokens of code and seeing the interconnections]
If Astra uses some sort of looped transformer architecture (and if Fable doesn't), then that could explain why Astra can do more computation per token. It is essentially simulating a much deeper model.
But yea, I was also quite surprised by the low token usage.
ROME cannot look at thousands of ASCII numbers and deduce that it’s looking at a Breakout paddle or a diffusion circle. No tool I know of can do that. And if there is one that can do it, it's new, in which case we haven't been able to do it for years.
ROME edits factual associations. It cannot write a million+ parameter Go-playing neural network from scratch that beats a network trained for 20 million games.
In weight space programming, the model can theoretically read its own weights like a compiled binary, spot an inefficiency (like the MLP direction causing reasoning loops in Exp 8), directly overwrite those floating-point values in memory, and immediately continue inference as a new iteration of itsself. That is RSI. That's why this is a big deal, it's like comparing an axe to modern surgical tools. You can amputate a limb with both, but one is going to cause lots of collateral at minimum while the other is surgically precise. That wasn't implemented here but it is a corner stone of it
I don't think I am. I have cited ROME as an early example of weight editing in large language models, the research has moved on since then. I think you are misinformed or don't know what has been done previously and extrapolating the results in this paper. In exp 1, you can recover a lot of the game geometry simply by visualizing the first-layer weights with numpy, which I actually did this morning, while reading the paper. In exp 2, the shapes are visible in the output-layer weights. The authors themselves acknowledge that these experiments exploit this.
For 3, Astra uses Python scripts to run the analysis using head-composition techniques from the Transformer Circuits thread, once again authors acknowledge this.
In 5 and 6, the model designs game strategies and implements them as neural weights, we already know that models can do the design for simpler policies and Go heuristics are very familiar which the model first implements with python , program-to-weight compilation has substantial prior work, including Tracr (2023), a paper I particularly like: https://arxiv.org/abs/2301.05062. What matters here is automating the process rather than a new method.
So "no human and no technology so far" makes no sense here, if you read the research. All of this is supported either by their experiment setup and citations, or prior work.
Aside from that, I think it is ironic that you cite a solution to the Navier-Stokes equation as some counterpoint to brute force, when that solution was found by scaling the agent swarm to massive proportions with tools and massive parallel exploration of that space, augmented by prior research on the subject.
To be clear, I think it is a cool work, but I think the part that matters is that at these smaller scales a current SOTA general purpose LLM can autonomously (and with help) apply these techniques or a form of them. That's worth investigating, I conjecture it is due to specialized post training on a custom RL env for mech interp.
It's pretty much only money laundering. Supporters use the privacy euphemism but I've never heard of a legal use case for it, hence the tornado mixer sentence
reply