>What I find fascinating is the application of information that is not present in the data but that can be confirmed by the data. I’m thinking specifically of deblurring algorithms people use to recover text.
I would argue that high frequency information is not missing in a blurred image, rather it's spread out spatially. Every sample in the blurred image is a convolution of multiple samples (i.e. high frequency data) from the originating source. Theoretically, it should be invertible, provided the exact function is known that does the convolution, but there are practical limitations, such as knowing how to apply the inverse operations at the edges and dealing with the extreme sensitivity to errors.
> And moreover, something already mostly solved algorithmically. Eg in zotero you can drag and drop files and it generates a reference based on the metadata as long as that is enough
As mentoned somewhere else, zotero's metadata retrieval is like 50% hit/miss in my experience - as in for the papers i've imported it was able to ID about half of them. An the quality of that data is not that great either. Lots of OCR errors, sparse bibtext fields, author names wrong, etc.
I'm a graphics programmer, i keep a lot of journals for research, and i needed the least painless method to get them sorted. Even if, say 1% is misclassified or has wrong metadata, it will not impact what I do the least bit. And those can be corrected with ease.
>At least with Zotero Version 6, you could have just selected all your PDFs and drag-n-dropped them into the interface.
I'm on version 9. I did the drag and drop for some of the files, but Zotero has only like 50% hit rate with finding the correct bibtex, and those are not that high quality either. People talk about hallucinations using LLMs, but I found them to be more reliable and produces richer metadata than Zotero's system. You can even use the Beaver plugin to correct the metadata, if you find them dubious or low quality.
>All it takes is a single hallucinated reference, and your career would be forever tainted.
Of any references I use, I always read the PDFs. The entire point of me using Zotero is to manage PDFs and being able to read them. If the title and author data in the item view don't match the PDFs then it's easy to notice. And even if i get the reference wrong, i'm not worried about my career, because it wouldn't matter.
I just finished importing and sorting thousands of PDF research papers, about 50 GB I've accumulated over the decades. Originally, I was sorting them in file system directly, but the whole thing got out of hand at some point, and thus the Documents folder became a dumping ground for PDFs. So I decided to take control of the situation and use Zotero to manage all this stuff. First I used Claude Code to generate all the BibTex files for the PDFs, then sort them into categories (still on the filesystem). Then created a master bibtex file that lists all the docs in each category. I used that to bulk import in Zotero, which then normalised all the file names once imported. What's really cool is that you can use Zotero built in web APIs to get an external tool to manipulate your collection directly. I've also setup WebDAV to sync the entire collection across multiple machines. Can you do all these without Zotero? Absolutely. And I did, but can't say I'll be looking back any time soon.
Not quite true. Lots of Thai dishes use a tonne of fish sauce and even shrimp paste in their dishes. They even make side dish dipping sauce (Nam Jim Jaew) that's like basically 50% fish sauce.
Agree in principle. But people like her does not have the resources, financially and emotionally to go through the legal system again. Unless there are charitable lawyers who are willing to do it on her behalf for free.
Yeah I can see that being a source of annoyance for situations like yours. However, I welcome it from a privacy perspective. The indicator alerts the user if some nefarious application covertly enables the microphone.
> If you have to explain it, there is zero chance of massive adoption.
Here's the thing, one should not need to explain it no mire. Devices or applications accessing content with an RSS option should present it to the end user through a convenient interface.
I would argue that high frequency information is not missing in a blurred image, rather it's spread out spatially. Every sample in the blurred image is a convolution of multiple samples (i.e. high frequency data) from the originating source. Theoretically, it should be invertible, provided the exact function is known that does the convolution, but there are practical limitations, such as knowing how to apply the inverse operations at the edges and dealing with the extreme sensitivity to errors.