There is one mention of Minecraft in the second paragraph of the Architecture section, "...We train on a subset of open-source Minecraft video data collected by OpenAI[9]." I can't say whether this was added after your comment.
The paper isn't about splitting water to yield hydrogen and oxygen gas which would be useful for energy applications. It's about a new way to make radical hydrogen (1 protein plus 1 electron) which is useful for synthesizing some organic compounds. It will be helpful for synthetic chemists and will make it easier to explore hydrogen radical chemistry. It may replace some processes that currently require transition metal catalysts, especially Samarium which is a rare earth element.
Anything that creates hydrogen radicals could conceivably be used to construct electricity-producing fuel cells.
Fuel cells normally radicalize hydrogen by contacting hydrogen with platinum catalysts under extreme conditions, requiring expensive and tricky design.
If this new process can do the same in mild conditions with inexpensive organic catalysts under exposure to light, it could lead to more economical fuel cell designs.
If I'm not mistaken this doesn't appear to be a catalyst. The reaction is driven by the irreversible formation of a strong P=O double bond, and the reagents are consumed. So great for chemical synthesis, but no clear path to fuel cell type reactions.
I did not look at the source paper before posting, but the paper makes it clear you're right.
However, the Wikipedia page titled "Phosphine Oxide" states that reduction of the oxide back to its original state is straightforwardly done with cheap reagents.
“In chemistry, a radical, also known as a free radical, is an atom, molecule, or ion that has at least one unpaired valence electron. With some exceptions, these unpaired electrons make radicals highly chemically reactive. Many radicals spontaneously dimerize. Most organic radicals have short lifetimes.”
Not an expert, but I did read the article. The point of the "Case-Shiller" metric is that it is computed on the basis of the same exact property changing hands in different time periods. That way it controls for trends in the product pool.
Or as the article put it:
Case-Shiller requires two transactions for the same house,” Lazzara said... It controls for the variability in the quality and size of the homes sold from year to year by measuring the change between houses that sold in one period with the prices of the same houses when they last changed owners. “The repeat sales mechanism is a way of adjusting for the mix of product so that you really are getting an apples-to-apples comparison,” Lazzara said.
In the case of lossless files, the takeout files are empathically not the same files that were uploaded. Google Music would allow a user to upload lossless FLAC files, but internally it converted them to 320 kpbs MP3 files. So, GPM certainly transcoded a portion of uploaded files. I'm not sure to what extent it left files alone if they met Google's formatting specifications. Perhaps someone else knows.
I don't think they did very much leaving things alone. One of my biggest problems with GPM was that my uploads would seemingly get de-duplicated alongside some other record that wasn't exactly the same, like a reissue or a remaster of the same record that sounded noticeably different. Sometimes an album I uploaded would gain a mysterious bonus track. They also at some point hosed up the whole system in such a way that many of my records contained every track twice, which meant I had to make playlists out of my old albums just to remove the even-numbered tracks and make it listenable again.
If you takeout from YTM it says your music files are "Your originally uploaded audio file" which is nice. Since music in YTM may have been migrated from GPM, that seems to imply that GPM retained the originals.
When they shut down GPM I migrated to YTM, which doesn't seem to have these specific catalog problems. I also just re-organized my local copy of my FLACs using MusicBrainz Picard. Unlike this author I no longer have the giant wall of CDs!
I remember enjoying E.P. Thompson's take on "Time, Work-Discipline, and Industrial Capitalism" in college. He has a lot of interesting commentary about how technology in the form of accurate timepieces played a role in our concept of labor. The article is here behind a paywall (https://www.jstor.org/stable/649749). Anyone with access to a search engine can likely find a free copy ;).
I think this counts a double blind trial. From the top of page 4 in the article:
"Randomization was performed through a computer-generated list stratified by site. Treatments were assigned after confirming the correctness of the admission criteria. Neither the research performers nor the patients were aware of the treatment assignments."
> As I understand it, this article is about a controlled experiment. In which case there is no confounding and therefore correlation implies causation.
Without a testable, falsifiable theory -- an explanation, correlations don't --
can't -- imply a cause-effect relationship. Not in a scientific sense, anyway. If I say that puddles cause rain, people will laugh. But if I say I performed a controlled experiment in a very large building like the VAB at Cape Canaveral in which I produce puddles and the puddles really do cause the subsequent rain, but I omit the details (or simply don't understand my own result), I get the last laugh. It's true, but it's not science unless there's a theoretical dimension, not just a description. It's not science unless I understand my result.
It's worth noting that convex.jl is developed by the same folks as cvx (matlab) and cvxpy (python). They all provide access to the same solvers with similar abstraction and usability.
