Fernand Pajot

@SanzoFilm @Dexerto Would a modern British accent which is actually farther away from historical accents than an American accent make it more relatable?

@Dexerto using american accents for ancient greek characters because it's "relatable" is the same logic that's been flattening historical films for decades, relatable and authentic are not the same thing and nolan of all directors should know the difference

Christopher Nolan explained why characters in The Odyssey have American accents "When you read the poem, it’s very earthy and accessible ... what I wanted to do with this film was really take a modern audience and throw them into a very relatable world"

@RichardHanania @sethharpesq Also it's trivial to apply a "film" filter to the iPhone image, which has much more raw information.

@sethharpesq Thank you for the insight into socialist logic.

Technology is making the world worse. Nothing good has been invented in the last 25 years. Superfluous, spurious innovations that seem impressive at first, like the iPhone, take back more than they give and ultimately benefit only oppressive governments and their oligarch allies.

@cmuratori The Xbox is a good Blu Ray player though…

@cmuratori I got disillusioned by “physical media” when I was living in the mountains with a 3Mbps connection in 2022 and thought “I’ll get an Xbox Series X and just buy game discs!”. The first two games discs I got greeted me with “you must download 80 GB to play the game”.

Apropos of the recent "no more physical media" stuff, I just wanted to mention that a year ago I unsubscribed from several streaming services and started buying Blu-rays again. I have been very happy about this decision.

@patrickc Have you read Meetings With Remarkable Manuscripts?

I've been enjoying Victoria Whitworth's new work, The Book of Kells: Unlocking the Enigma. I've actually never seen the Book of Kells in person, somewhat to my embarrassment. I've been doing some reading about the origins of Christianity this year, however, and I figured I should know something about the most famous Irish manuscript. (Perhaps the most famous manuscript, full stop.) Reading the book, I was struck by how much the contents have suffered over the past ~1200 years (enduring everything from water damage to reckless malfeasance in attempted nineteenth century restoration), and I wondered whether AI could help give a sense for how the work might originally have appeared. I downloaded the Internet Archive's PDF and asked my friendly neighborhood agent to use gpt-image-2 to render each page the way it imagines it might have originally appeared. Remarkably, this all worked with a single prompt, with the agent spinning up 48 workers, since each page took a minute or two. (I'm sure that someone wiser than me could prompt the model better, ensuring somewhat more historical accuracy in color restoration and so forth. There is no gold leaf in the Book of Kells!) This part of the project went from conception to completion before I'd finished my morning coffee. I then wanted some easy way to view the results online, so I asked Stripe Projects (projects.dev) to host the result on Vercel. That also worked in basically a single prompt: bookofkells.vercel.app. I also figured that people might want an easy way to download the full PDF of updated images, but it's a large (~200MB) file, so I decided that I should charge $0.10 to cover bandwidth costs using @MPP. I asked my agent to set this up, and it basically worked smoothly, though I had to tell it what MPP is (I guess it's not yet in the pretrain) and also manually set up the Cloudflare account that actually hosts the PDF and configure the API key. (Vercel seemingly has a 100MB limit.) The purchases now show up in my Stripe account alongside all other activity. The site now has a ready-made agent prompt for anyone who wants to download the whole thing. I'm guessing that we'll see a lot more UIs like this in the future. I remain pretty intrigued by the intersection of agents, micropayments, and stablecoins. I don't know much about managing crypto wallets from the CLI, but now AI can do that for me, while Stripe seamlessly handles turning it all back into fiat. So what is the moral of the story? • Whitworth's book is very good, and you should buy it. • The Internet Archive continues to be wonderful and a civilizational treasure. • While there are rough edges, setting up third-party services via the CLI now basically works. I'm pretty sure I wouldn't have bothered with any of this if I couldn't have outsourced almost all of the work to AI. • The image models have gotten very good. • There will probably continue to be all kinds of interesting applications of AI to history. (The Vesuvius Challenge of course being a shining pioneer.) • These days, I often find myself building single-use sites for things I'm learning or for books I'm reading. I think this is a cool new category of software.

@antovsky @nicbarkeragain @wookash_podcast Have you played Gladiabots? gladiabots.com

Come see a sneak peek of the game @nicbarkeragain and I have been working on! We're also doing a live podcast with @wookash_podcast about it tomorrow :) If you have any questions about the game/tech just reply here, and we can try to answer live on stream 🫡

@lauriewired I'm not sure I understand, CPU and GPU caches have been paramount for the last 10+ years? Also most GPU kernels have been about efficient L1 <> SMEM <> VRAM pipelines.

funny how the best bang for the buck in CS research is having a thorough understanding of heterogenous memory / tiering again last time that happened was the ~60s

@sebkrier It’s a little bit fluffed / drawn out out, just a heads up

@fernandp Will do 🫡

🥺

@__tinygrad__ CuTe

Today, if you were writing a bunch of kernels, what would you reach for? Raw CUDA? tile-lang? Triton? ThunderKittens?

@ThePrimeagen @jakedowns It depends on what you measure, I they store wave data instead of converted pixel data, and keep it at the canonical resolution that can be a lot for volumes.

yeah its something like 40gb per image or 17gb / s of data flowing still don't quite understand it, but at the end of the day its an insane amount of data and i have SOO many questions I just assume that its 1GB per image (super high res for medical quality). that number is insane

the midjourney pivot is so much crazier than you realize

@ThePrimeagen @jakedowns Btw they don't store 800 TB per scan, that's just the throughput before you aggregate. Similar to CT/MRI for that (though probably still more data).

the word expansion is doing heavy lifting here image generation -> body measurement via water vibrations incredible infrastructure that is literally alien today and a spa! several of them actually class image generation -> making ~800TB per scan -> processing that into a much more compact version on servers -> storing it -> retrieving it -> a spa....

@Love2Code With different hardware than matrix mul centered hardware, probably.

@fernandp IMO we can likely discover architectures that are much more efficient than transformers. Potential for huge gains there, but there's not that much research in that area because it's risky to try something that might not work. It's still going to happen eventually though

IMO RAM prices will come down, because the datacenter buildouts rush is hitting a wall with respect to permitting and power usage. The next focus is not going to be hyperscaling, it's going to be optimizing both training and inference for efficiency... And it's about f*ng time.

@Love2Code We'll keep hill climbing this for sure, but I think we're already doing really well there.

@fernandp Much better efficiency is possible on every front. It will seem obvious in hindsight.

@scaling01 I don’t think you understand Xi Jinping’s values. Directionally yes they probably will increase investment over time.

Prediction time: - Europe will absolutely learn nothing from this incident. At best they are going to invest an additional 2B Euros in some robotics or image gen startup, because they think that's somehow relevant - China will get even more AGI pilled and restructure their entire economy for one purpose

@cmuratori Something interesting to think about: say that at some point in the future a model is not trained on any pirated stuff (everything is properly licensed etc), but is also trained on synthetic data from older models which were trained on pirated stuff.

For people using AI in commercial game development: I'd be interested in hearing the best arguments as to why you think people should pay for the resulting game instead of pirating it. Concisely, if you pirated the inputs, why shouldn't they pirate the output?

@nabeelqu They bio safeguards have always been overly strict :(

They were not kidding about overly broad safeguards...

So far, Fable is absolutely crushing my private evals. Excellent model.

@alerityy @ladybirdbrowser Ahem...sqlite

@ladybirdbrowser what a horrible change on so many levels - defeats a lot of the benefit of FOSS IMO, and will significantly hurt this project going forward. feels like this post was almost written as if you guys forgot the concept of simply… reviewing PRs before merging them? 🤯

Ladybird is moving into a new phase as we work toward our first alpha release. We are tightening how code enters the project: going forward, code changes will only be introduced by project maintainers, and we will no longer accept public pull requests. ladybird.org/posts/changing…

@avivafashion11 You forgot that they make you fill even more forms on paper when you arrive

Can someone please explain this to me?? The doctor’s office will send 17 reminders, 4 emails, 3 texts, and a carrier pigeon reminding you to arrive 15 minutes early for your scheduled appointment. You fill out every form online in advance and dutifully arrive 15 minutes early to check. Receptionist: “Have a seat.” 45 minutes later, you’re finally called back. 🫩 Why are only one of us expected to keep the appointment time?!