It used to take years, decades, or centuries before a system could grow and evolve to be so complex and unwieldy, and so full of internal contradictions, that the whole thing becomes an incomprehensible tangle of hairballs. An example is the patchwork system of international, national, regional, and local laws we have at present, which has grown and evolved over centuries.
It's a worthwhile effort. If successful, Woxi can enable a large mass of scientists and engineers who don't have access to Mathematica to run legacy code written for it. Also, Woxi would give those scientists and engineers who regularly use Mathematica a non-proprietary, less restrictive alternative, which many of them would welcome.
How does Woxi compare to other "clean-room implementations"[a] of the same language?
--
[a] Please check with a lawyer to make sure you won't run into legal or copyright issues.
The full quote doesn't fit as a headline: "I used to think that if there was reincarnation, I wanted to come back as the President or the Pope or as a 400 baseball hitter. But now I would like to come back as the bond market. You can intimidate everybody." - James Carville
Yeah, this is a much clearer source and the abstract gets pretty directly to the point. The first paragraph tells you pretty much everything you need to know before you read more. The Ars article took 4 paragraphs to mention "client isolation" and even longer to get into the meat.
At first glance, this looks incredible to me. The authors train one model on 40K hours of computer-use video, previously labeled by contractors with keyboard and mouse actions, then use that model, in effect, to label 11M hours of computer-use video, which they use to train the computer-action model. The key advance is in compression. Quoting from the OP:
> [previous models] burn a million tokens to understand just one minute of 30 FPS computer data. Our video encoder encodes nearly 2 hours of video in the same number of tokens—that’s 50x more token-efficient than the previous state-of-the-art and 100x more token-efficient than OpenAI’s encoder.
While I was already aware that there are people working on new, more efficient "world models," this is the first one I've seen in action. I'm a bit in shock at how good it is, quite frankly.
I've added the OP, as well as a related 2018 paper on Behavioral Cloning from Obervation (BCO) to my reading list.[a] So far, I've only skimmed the 2018 paper, but it's already evident that it's well-written. I'm no expert in deep RL, and I can understand it. BTW, "Behavioral Cloning from Obervation" is a really good name, with an easy-to-remember acronym.
yeah! i love the BCO paper, i think its extremely intuitive and these methods are really interesting in a time where data without labels is abundant. i especially like the idea of iteratively making the inverse dynamics better—might lean closer to that in the future
> i especially like the idea of iteratively making the inverse dynamics better
Same here.
The notion of inducing these models to "hypothesize" distributions over possible actions given subsequent observed transitions makes me think of "contrastive divergence," the method Hinton and others came up with for unsupervised training of Restricted Boltzmann Machines (RBMs), in the prehistoric era of deep learning.
Given each training sample, an RBM would 1) execute a forward pass, 2) sample its output units, 3) "hypothesize" its input units, 4) execute another forward pass on the "hypothesized" input units to sample new output units, and (5) compute a type of contrastive error for local backpropagation. RMBs could be stacked, with output units from one becoming input units for the next one. Hinton called the input units "visible," and the output ones "hidden."
It's not the same, obviously, but the idea of modeling machine-generated inputs (or actions) given outputs (or transitions) has always been appealing. It has a long history.
Sigh. Anecdotally, more Europeans no longer want their governments to rely on software and data controlled by US companies, because they no longer trust the US to act as a reliable ally, defending the same values. Whether you agree or disagree with these concerns, they are valid for many Europeans.
In an ironic twist of fate, the US government's actions could end up causing long-term damage to US tech companies.
This is all based on anecdotal evidence, so I could be wrong, but I have to call it like I see it.
At the end of the day it isn't US tech companies that'd suffer (outside some minor short term pain) it's the US. If being in America is bad for business those companies (which already exist multi-nationally in most cases) will just pack up their US holdings.
Thank you for sharing it on HN!
reply