mkaic

If you want to mess around with these sorts of visualizations yourself, I recommend checking out David Bau's little web app for it: http://davidbau.com/conformal

Anecdotally, I've generally heard this phrased as "for better or for worse" :)

I was referring to link [1] in GP’s comment, which is to a paper published in 2021, not to either of the more recent papers they published.

Can't believe that FNet paper flew under my radar all this time—what a cool idea! It's remarkable to me that it works so well considering I haven't heard anyone mention it before! Do you know if any follow-up work was done?

I am thoroughly confused about how this ad has become the controversy it has become. This whole ordeal feels very "let's invent something to get mad at!"

I would bet a large sum of money that the creatives who created the ad for Apple had no intentions beyond "this ad will be a metaphor for the insane variety of creative endeavors you can use an iPad for!"

Apple isn't tone-deaf. The people mad about this are tone-hallucinating.

> We believe developers and users should have the flexibility to use our services as they see fit, so long as they comply with our usage policies. We're exploring whether we can responsibly provide the ability to generate NSFW content in age-appropriate contexts through the API and ChatGPT. We look forward to better understanding user and societal expectations of model behavior in this area.

Seems even OpenAI can't resist the massive amount of money to be made in autogenerated smut. They've probably seen the huge popularity of their less "morally scrupulous" competitors and decided they want a piece of that pie.

The YouTube channel Hyperspace Pirate[0] has been consistently uploading fantastic videos documenting their attempts to make their own homemade liquid nitrogen machine. They don't shy away from including all the technical details, and throw in plenty of hilarious deadpan humor too. Highly recommend them if you're interested in this topic.

[0]https://www.youtube.com/@HyperspacePirate

Fascinating. I am only familiar with this definition of the word! To me, "nonplussed" has always meant "unimpressed, indifferent", adjacent (but not quite identical) in meaning to "nonchalant".

I love watching language evolve before my very eyes, it's literally the coolest thing ever!

I think calling them a "con artist" is pretty over the top, no? To me, this seems like a clear case of a more junior game dev being very excited about a cool thing they built and optimized. Sure, it may not really be the holy grail they think it is, but they aren't doing any harm by being excited of it and proud of themself!

Exactly. It's like having a bazillion tiny 3D printer extruder nozzles embedded in your skin all on their own timers and flowrates.

Exactly—getting it that high is one of the primary objectives of the Starship initiative.

IIUC, it doesn't. Hair grows for a certain time period, falls out, and regrows. The follicle has no knowledge of the length of the hair it is producing, it is simply on a grow-stop-shed cycle with a certain period.

Of course, a discrete, finite sampling of a square wave at a set of points in time only requires a finite number of coefficients to perfectly reconstruct.

> AI is opaque... The amount of papers I’d have to read and mathematics that I’d have to ingest to really understand why a certain prompt X results in a certain output Y feels overwhelming.

Modern deep learning is almost wholly built off of partial derivatives, the chain rule, and matrix multiplication. It looks complicated from the outside because there are so many people publishing variations on this same basic formula, but it is the same basic formula over and over and over again. I may be a "new fart" myself, but when I set out to become an AI researcher I was honestly kind of surprised how transparent the mechanisms underlying AI actually are! I had expected to need to spend years to learn what I needed to understand it, but in reality it took me about 6 months of dedicated study to get to a point where I felt more-or-less comfortable with how it all worked. Granted, learning how Transformers and attention worked specifically took me a little longer (not sure why, was just some kind of mental block)

> AI is Not Approachable

I would argue that AI is more approachable than it ever has been. A used RTX 3090 with 24GB of VRAM can get you shockingly far for ~800 bucks. Will you be training GPT-5 at home? No. But there is still so much fascinating fundamental research waiting to be done in this field. Personally, I really enjoy using my home PC to implement and train funky novel architecture designs and see if they work on common toy problems. My hope is that one day I'll stumble onto one that becomes a Transformer-killer and then I could get a compute grant or something to try scaling it, but that's likely a pipe dream. Still, I find the amount of cool stuff you can do with consumer hardware today astounding.

> AI is Not Open

At the risk of sounding like a broken record, I think AI is more open than it ever has been. Yes, the state of the art is closed-source right now, but Open Source AI has never been as active and exciting as it is right now. Mistral LLMs, CLIP, Whisper, TheBloke and his crazy quantizations, DINOv2, LLaMA-2, Stable Diffusion, and so much more have been released open in just the past year or two. Local LLMs are slowly (but surely) catching up to GPT-4, even if they are lagging behind by about 6-8 months. The best open LLMs these days are (anecdotally) above the level of ChatGPT-3.5 when it launched. That's exciting! Of course all the coolest, most sexy AI research is happening behind closed doors, but to see that and conclude that AI is "not open" feels shortsighted.

In conclusion: What a time to be alive! I think it's perfectly valid to feel nervous about the future, because things, they are a-changing! And change is scary! But I really do believe that there is more to be excited about than there is to be worried about, and that everything will Work Out™ in the end.

I think underground metros, while expensive to build, solve this, because they are always more convenient compared to any traffic-ridden area. A bus can get stuck in the same traffic as the cars, and bikes taking over any significant portion of American commuting in the near-term feels laughable to me — we are quite culturally different from the Dutch :P

Article is a bit of a nothingburger, here's the quote in question from Zuck's Instagram:

"Some updates on our AI efforts. Our long term vision is to build general intelligence, open source it responsibly, and make it widely available so everyone can benefit. We're bringing our two major AI research efforts (FAIR and GenAI) closer together to support this. We're currently training our next-gen model Llama 3, and we're building massive compute infrastructure to support our future roadmap, including 350k H100s by the end of this year -- and overall almost 600k H100s equivalents of compute if you include other GPUs."

It's not common for companies to have many breaks at all in the US — paid time off is a rare luxury for many workers here. The country's culture is riddled with Protestant views on work and its virtue, which when combined with our rampant individualism and disdain for public works leads to The Hustle being an idolized state.

It has its upsides and downsides. Upsides include the country generally being pretty darn industrious and a hotbed for entrepreneurship, and having good socioeconomic mobility prospects for those who (A) don't start out genuinely poor, and (B) are willing to be obsessive about work. Downsides include burnout, dispassionate employees, and an icky tendency to associate poverty with immorality and wealth with morality. I personally believe we could definitely stand to move a little further in the European direction when it comes to labor culture and laws here in the USA.

Prescriptivism vs. descriptivism. Taxonomically, you're right — kelp is a macroalgae, not a plant. But most people don't really care (and don't need to care) about this taxonomical distinction in their day-to-day life. It has a stem and leaves, it eats via photosynthesis, it's rooted to a single spot, and it's green-ish — as far as I'm concerned, that's a plant!

I generally find that defining words descriptively (based on how most people use them most of the time) rather than prescriptively (based on how institutions or niche groups use them) is a good way to avoid conflict, both on the internet and IRL!

San Francisco also refuses to enforce basic rule of law and has hamstringed its police department, so this doesn't surprise me. Crime rates are through the roof, and a housing crisis has been ongoing for years because of restrictive zoning. I don't think it's fair to assume that basic income for homeless folks is the cause of SF's devolution into a hellhole. There are a lot of other confounding factors.

You're totally right, apologies for my non-answer earlier :)

I feel like education is probably our best bet for sort-of-kind-of dealing with the repercussions of the problem. Ultimately, attempting to legislatively regulate the production and private distribution of AI nudes just doesn't seem enforceable to me — see: Brandolini's principle. I think our best hope to mitigate the negative impact of this tech is to talk widely about it and actively try to make my "prophecy" come true. Sex ed curriculums should start including warnings about how fake nudes are an epidemic and begin to build the new social norm of not assigning moral judgements based off of them.

I could also see some government-funded PSAs on popular streaming services/TV channels/radio channels being effective. You'd want to try to disproportionately target such messaging to older members of society, since they're less technologically literate on average and therefore more likely to be emotionally traumatized by tech like this.

Now that I'm writing this, I do wonder if there are some regulatory approaches that could help, especially since we have had some success banning CSAM legislatively. The problem with AI nudes, I think, is that they are destined to become indistinguishable from genuine nudes, so a blanket ban is impossible without establishing a panopticon (which is obviously bad). Blanket banning works for CSAM because there is never any case in which a piece of CSAM should ever be knowingly stored or distributed, regardless of its origin. It is always immoral. This complicates things because nude images of adults can be shared consensually by said adults, and banning this practice would be absurd.

I'd be curious to hear your thoughts as well. What sorts of options are there to combat this new AI-generated nude plague if my optimism about the situation turns out to be misplaced?

With the honestly astonishing ramp up in solar power and grid-scale battery production over the past decade, I have to disagree with this take. We're hurtling towards solving climate change through market forces alone — now, that doesn't mean we shouldn't also be using policy to speed the solution, but it is pretty clear that, as Noah Smith puts it, "Solar and batteries are going to win, and our thinking needs to adjust accordingly."[0]

The coolest statistic I've heard regarding solar and batteries is that 0.5% of worldwide GDP is being invested in it. One out of every 200 dollars spent is being spent on solar energy! It's truly incredible :)

[0]https://www.noahpinion.blog/p/our-climate-change-debates-are...

I mean, I'm just making an educated guess based on how society has historically been quite good at changing its norms based on new technology. We didn't have cars, and then we did, and within a generation it became extremely normal and expected for people to drive. Digital cameras were a big deal when they were invented, but for people like me who grew up in the early 2000s, they have never felt that unusual because I've been exposed to videos of myself as a baby for as long as I can remember. When The Avengers was released, it caused a huge amount of hype and hubbub. Now, Marvel produces a dozen superhero movies a year and nobody cares anymore.

Do I think that society will transform its views on nude images overnight? No, obviously not. Do I think that every person in society will go with the flow and adopt new assumptions about nude images? Also no. But it seems entirely logical to me to imagine that once the average person notices "hey, I am being inundated with 1000x the number of fake nudes that I used to," they will eventually reach the conclusion that "I should start assuming all nudes are fake by default."

Agreed. The "transition period" between this technology being invented and society finally adapting to its existence is probably going to be quite rough for a lot of people. My only hope is that society adapts quickly.

The point isn't that it wasn't possible before, it's that it's 100x faster, easier, and cheaper to fake nudes now than it was previously. I think the reason for why claims of leaked nudes being fake are scoffed at is because making 100% believable fakes takes non-negligible time and effort, and maybe even money. Now, it's just "upload clothes-on image, download clothes-off image 20 seconds later". I think it's not insane to expect this huge shift in accessibility to eventually cause social changes — in the same way that asking for someone's Instagram is normal now, despite Instagram being a relatively new invention, I think it's pretty obvious that a few years from now, nude photos will be assumed by default to be fake rather than real. The social standards are a product of reality, they don't determine reality.

And this is why I began my comment with "I can't speak for everyone!" Art is so subjective that it would be rather silly of me to try to make a bold, authoritative statement about its nature. I definitely understand your perspective to some degree, and wish you the best :)

I can't speak for everyone, but the reason I personally find generated images completely uninteresting from an artistic standpoint is that they are low-effort. I appreciate art because a human sacrificed time and effort to make it. As a general rule of thumb (with exceptions, of course), the more time and effort an artist sacrifices to make a piece of art, the more I find that art interesting, especially if the artist is highly skilled.

I don't think that AI art is fundamentally uninteresting (I am a machine learning researcher, so I think it's rad), or that it can't be used as a tool to make great art. I just think that the final image is not the point of art. The point is that a person practiced for hundreds/thousands of hours to develop a creative skill and then spent several more hours painstakingly applying that skill just to create the image. If AI is incorporated into a high-effort, high-skill process, I don't have an issue with it.

Have heard from others attempting it that it needs 40GB, so basically an A100/A6000/H100 or other large card. Or an Apple Silicon Mac with a bunch of unified memory, I guess.

I don't think transformers or diffusion models are inherently "unsupervised", especially not the way they're used in Stable Diffusion and related models (which are very much trained in a supervised fashion). I agree with the rest of your points though.

Almost certainly. As a longtime member of Ian's Patreon, I recognize more than a few specific techniques being used here that he's posted tutorials for on it. Not that the inspiration takes away at all from the final product! The fact that a 16yo made this would be insanely cool even if it was a shot-for-shot remake of an existing movie scene, so any originality that it has beyond that only makes it more cool!

I want more immigration AND more fertility, personally.