Ha, my previous comment was before your new edit mentioning Sora. There is a good reason why the accompanying research report to the Sora demo isn't titled "Awesome Generative Video," but references world models. The interesting feature is how many apparently (approximations to) physical properties emerge (object permanence, linear motion, partially elastic collisions, as well as many of the elements of grammar of film), and which do not (notably material properties of solid and fluids, creation of objects from nothing, etc.)
Yeah, I've been thinking about similar concepts in a different context. Fascinating.
Regarding the role of time, the idea of a purely conserved quantity is that it is conserved under the conditions of the system (that's why the article frequently references Newton's First Law), so they're generally held "for all time that these symmetries exist in the system".
Specifically on time: the invariant for systems that exhibit continuous time symmetries (i.e. you move a little bit forward or backward in time and the system looks exactly the same) is energy.
Here's my ELI5 attempt of the time/energy relation:
imagine a spring at rest (not moving)
strike the spring, it's now oscillating
the system now contains energy like a battery
what is energy? it's stored work potential
the battery is storing the energy, which can then be taken out at some future time
the spring is transporting the energy through time
in fact how do we measure time? with clocks. What's a clock? It's an oscillator. The energized spring is the clock. When system energy is zero, what is time even? There's no baseline against which to measure change when nothing is changing
There are many machine learning problems which should have symmetries: a picture of a cow rotated 135 degrees is still a picture of a cow, the meaning of spoken words shouldn't change with the audio level, etc. If they were doing machine learning on tracks from the LHC the system ought to take account of relativistic momentum and energy.
Can a model learn a symmetry? Or should a symmetry just be built into the model from the beginning?
Equivariant machine learning is a thing that people have tried... Tends to be expensive and slow, though, and imposes invariances that our model (a universal function approximator, recall) should just learn anyway: If you don't have enough pictures of upside down cows, just train a normal model with augmentations.
Could it be something like "failure is the mother of success"?
(失败乃成功之母)
I am no expert, I was curious myself and went down an internet-sleuthing rabbit-hole. Apparently this is a common phrase, but I am not fluent in Mandarin, so cum grano salis.
That sounds like a close cousin for sure. It’s hard to tell because it was in English but maybe the Chinese was in small script below. I don’t know if I still have that manual either, it may have been destroyed in one of these boxes that got left out in the rain. I’ll try to look for it next time I’m digging around the attic.
Yes, Mercury was very prominent a couple of weeks ago. Of course it's always close to the Sun, so perhaps your horizon viewing conditions have obscured it (I have to walk so that the trees and street lights are out of the way).
I've been trying to track the Pleiades for Matariki (Maori New Year), which are currently just above Venus.
I directed a short film in the late 90s and we used this as an 'event' at a party (minus the floating part). We removed the glass from a light bulb to act as a remote fuse. Acetylene was perfect, as it gives this wonderful 'crack,' much better than the LPG we first tried. It was Tim's idea.
Codex has the capacity to write decent Python code. A large part of getting great results from it is writing clear, well separated, prompts. Also taking it off streaming mode, to get best N results, and penalizing repetition, improves the output dramatically. Even better if you can give further hints (e.g. import pandas) before submitting.
Of course, it's merely trained on code it's seen on GitHub, so it certainly has a particular smell to it (disclaimer: I focus on Data Science related code, which is not always of the highest quality and has its share of cargo-culting).
Most of the demos you will likely have seen are in streaming mode, have vague prompts with a high temperature setting.
What a great video. I appreciate the care and effort that went into this.
If you are interested in growing some oyster mushrooms, you can pick up a self-contained growing kit for ~US$20. It's not difficult, and it really is a lot of fun to mist your babies and watch them grow over your sink. They really are delicious simply sautéed in some butter and garlic.
For a more "hacker" experience with mycology I highly recommend starting from a spore syringe/spore print, but doing every part of the process yourself. You can grow any of a wide variety of mushrooms and it's really cool to understand the details of creating a sterile environment for the spores, creating the substrate, and following a wide range of "Teks" for varying difficulty.
Most of the online documentation is coming from people aiming to grow p. cubensis (aka magic mushrooms), but the techniques will work equally well with many food varieties (especially oyster).
The BRF (Brown Rice Flour) Tek is a good place to start, and you'll quickly begin to see how wild you can get with home mycology. Once you grow your first batch you can easily make your own spore prints, and start doing very fascinating stuff with home made agar (cloning, inspecting genetics, cleaning out contaminants, spawning spores, etc). There is also a fairly large range of options for automating the process with electronics if that's the kind of thing you enjoy.
If you get really experienced you can start taking spore prints of local native species and then growing them at home from spores! It's a far deeper hobby than I would have imagined, you can start from something very simple that just takes a few hours to start and can reasonably end up closer to something like this video than you can imagine!
Can confirm. The oyster mushroom kits sold are Lowe’s and Home Depot are fun, nearly foolproof, and much more delicious than any mushrooms I’ve bought at the grocery store.
Python's NumPy library deliberately emulates much of the functionality of Matlab's core (similarly, fundamental plotting with matplotlib follow similar design cues). As a lapsed long-time Matlab user, I think they're just as fun to use.
There is also Octave, an open-source Matlab alternative, but when I last used it (admittedly quite a while ago) it had limited support for toolboxes (e.g. signal processing). I imagine it has only improved since.
For elegant matrix syntax I like julia or fortran better, if I need a lack of a license or more speed. Octave is good too, but it is much slower than Matlab. Numpy is in a weird spot. Its still slower than Matlab, but its everywhere, but all the versions and environments are different so you need containers to be portable.
