It isn’t even remotely close. The heat capacity of water is 4.18 joules per gram degree Kelvin, almost nine times that of steel (0.466), and that doesn’t count the latent heat of fusion in the ice to water phase transition. Now, metallic lithium has a respectable heat capacity of 3.58, but I suspect the “Keep Kool Lithium Kube” experiment may end badly.
Burying the lede on the AI editing software he pitched at the beginning.
Re the editing software, what’s stopping a high school student from downloading the video of a virtual class and editing it to change what the teacher said?
From the brief demo, it looks like nothing. We are on the threshold of the great democratisation of deep fakes.
For readers who don’t know what we’re talking about, here is a promotional video for Descript, the sponsor of the video in the main post.
I think we need to slightly revise the Eagles’ song Lyin’ Eyes. Instead of "there ain’t no way to hide your lyin’ eyes, it needs to say “there ain’t no way to trust your lyin’ eyes (and ears).” The societal and legal implications of this technology seem hard to overstate. It causes the entrepreneur in me to consider the possibility of an authenticating method for videos, which can only be applied at the time it is created. I suppose blockchain technology (which I barely understand) lends itself to that important purpose. I’m thinking that “real”, “unedited” videos would bear something like the “Good Housekeeping” seal of approval (for those who are old enough to be aware of it) in every frame, so viewers could see it is authentic.
This tech will surely bias me against believing any video I see going forward - particularly those which show a non-progressive saying or doing ‘bad’ things.
Very nicely done promo video. It seems the software can clone the speaker’s voice, but it won’t necessarily deep fake the video? In the demo, it either showed taking out pieces, or overdubbing new content without having the speaker in the scene.
So perhaps we are not quite there yet in the realm of deepfakery? Maybe the new standard is “show my lips moving” in sync with the audio you hear on the video to rule out the sort of post-processing Descript allows.
There seem to be quite a few options for “hard” problems that AI can solve reasonably well. I recently used something called Vance.AI for image post-processing and it was quite effective quality wise and saved me time compared to doing it from scratch with Lightroom, Photoshop, etc. Another example is Otter.AI which provides automatic transcription at a reasonable cost. I would say these kind of examples provide good enough quality at a typically low price points. Curious to see where this will go next?