171–180 of 218 posts
Quite incredible work, with the original breakthrough model being trained on a 1070: https://twitter.com/LukeFarritor/status/1754532281690243339
Large Language Models have skewed the perception on the amount of compute required to do useful things with ML.
When I first came across this project on HN (early last year), I was taken aback by how impossible the project looked and how smart were people working on this. Despite seeing a few intelligent names behind the project, I subconsciously believed that this would at least take 5-10 years before a breakthrough. Today I sit with the same amazement, taken aback again, appreciating how ridiculously awesome this is. Congrat…
It's all about incentives. $1 million is a lot of money. The vast majority of hard problems don't have much brainpower dedicated to them, because the bang/buck ratio doesn't work out. Machine learning, math, and adjacent fields already have many careers that pay very well, so getting top-notch experts to dedicate their attention to what might be a futile endeavor is difficult. And this isn't only about the monetary v…
But it's not clear what current tech could help with. Machine learning can't be applied to something you don't have training data for.
Earlier quoted context omitted.
This is why I am going for cryopreservation if I ever have the luxury of choosing the way I die.
The part that I do not understand there is: why would anyone/everyone else want you to be alive again? And do not understand me wrong, I would very much be interested to talk to people from various ages. Specific ones, but also commoners. But why should someone want to restore thousands of random people from (e.g) 1284 on their own cost? That only works if those people have big stashes of money that are legally still…
Earlier quoted context omitted.
Funding is a huge one as well. Funding is the wheel that drives the project (source, have been hanging around the project people for a little while). If you know anyone that would help chip in for the Phase 2 of the project (scaling up, please let Nat know! (not directly affiliated with the project management team, just pointing to him as a great contact for that.... <3 :')))) ) )
It seems "weird" none of the mega rich has committed a few million dollars for this, it looks like a very good way to build a legacy while benefiting humanity, and e.g. Bezos would probably find a million dollars behind the couch pillows.
Herculaneum was one of the highlights of my trip to Italy with the wife. I didn't realize the scope of just how much ash and soil had to be removed for excavation. It was dozens of meters [1]. It's an absolute shame that the site is given a fraction of the attention that Pompeii receives, I thought it was vastly better preserved and truly awe-inspiring [2]. I highly recommend spending a few hours wandering the site,…
I was ridiculously excited when I first read about this in October (if I remember correctly) last year, when a few of the first results were beginning to pop out. I found the methodology fascinating. First of all the digital unwrapping of the scrolls, then the recognition that crackling in the paper was the sign of ink, and finally putting together a model to detect it, piece by piece. I need to look into the final r…
my understanding is that the scan they did on the scrolls returned the layers themselves. Like so: ``` xxxxxxxxxx xxxxxxxxxx ... xxxxxxxxxx ``` So, by tiling the image on the surface you get data that is size_x * size_y * n_layers. So, it can be seen as a video stream with size_x * size_y * 1 channel * n_layers where the layers replace the temporal dimension.
It's a cut through the scroll, with the time dimension in this video representing the location of the cut along the scroll lengthwise.
As you can see from the mess it's far from trivial to find the surface of any of the sheets in the scroll, often they layers are blended together messes.
You may be thinking of the scans they used of an unwrapped sheet, those were as you describe and were used to help figure out methods for the real challenge.
I think the biggest success isn't the recovered text but organizing such an endeavor with such success.
Totally inspiring!
Earlier quoted context omitted.
This is why I am going for cryopreservation if I ever have the luxury of choosing the way I die.
Well, that could go lots of ways. Maybe some rich trillionaire buys you and spawns you into an endless horror simulation. They might be into torture and get off on it. ("No real humans harmed.") But if the future can reverse the light cone, nobody is immune to that fate. Who knows what the future holds. These are just sci-fi flights of fancy.
An awesome achievement. I don't know if it's possible but I wonder if they could do energy-resolved CT and try to look for a differential absorption edge in that ink. I have no idea what material it is actually made of, but if the IR photo is different, then there's a nonzero chance they could generate quite significant contrast by hitting one of the K-shell edges of a characteristic ion (iron?) present in the ink bu…
Earlier quoted context omitted.
Right? Imagine trying to explain some of it to one of the ancients - so, you have this quartz sand, see?...
That would be like explaining a baking recipe by starting with protons and electrons. State-of-the-art machine learning architectures aren't actually that complex. Diffusion models and transformers can be explained to a bright high schooler. I'm sure Archimedes and Euclid would have no problem understanding them. What they might have a problem understanding (or even imagining) is the mind-boggling amount of computati…
It would be like explaining a baking recipe just talking about wheat and flour and heat. The point is that from first principles, ML is a huge jump. From first principles, baking is not.