This is awesome and very fast especially given all the data displayed!

This wouldn't have helped here but there is a related field of research called hotel recognition. Many of these videos are filmed in hotels so being able to recognize if it was a Mariott or even better a non chain local hotel can be very helpful to investigators. They basically train CNNs that learn to pick up on the bathroom fixtures or kind of bedding used by different hotels. One researcher in particular has done a tone of work on this: https://scholar.google.com/citations?user=mNoB9SgAAAAJ&hl=en

I wonder if you could get the interior of every house from Zillow/realtor websites and then do something like this for every house in the country... Clearview for bedrooms?

I think he's referring to CPU mitigations: https://en.wikipedia.org/wiki/Transient_execution_CPU_vulner...

Protip: Settings -> Personalization -> Base style and tone -> Efficient largely solves this for ChatGPT

The malleability of the ciphertext matters because it enables certain circuit tagging attacks as the article explains. It means that the exit relay could confirm you are using a guard relay also controlled by them and thus discover your origin IP address.

There are many reasons that these cryptographic tagging attacks are a lot worse than just the timing correlation attacks that are possible if you control the guard and exit of a client: https://archive.torproject.org/websites/lists.torproject.org...

Nice article and especially so for including the parsing that most people just outsource. What's great about using an emulator is that you can also do fun things with the syscalls like implementing your own "virtual filesystem" instead of just translating directly to the x86_64 equivalent syscall: https://github.com/gamozolabs/fuzz_with_emus/blob/master/src... (not my code but basically something like this)

OK, I think the problem was that you are only supposed to input the user ID number. I just limited the form to numbers only and updated the description to make this more clear.

Hi, I just realized that the confusion here was that you are only supposed to input the numeric user ID. I just limited the form to numbers only and updated the description to make this more clear.

Can you access your profile page in incognito (ie is your account public)? Alternatively, if you have more than 5000 books in your shelf, that might break it. I just tried a number of users and I was able to import them all.

Ah. My fault. I put the whole URL in the User ID box instead of just the User ID.

Do you know how your model compares to Cinematch's SVD? Does your model use only books rated highly or also include low rated ones?

I am not familiar with Cinematch, is there a writeup about it? When training I used every input book and did not include ratings as a feature. In the future I want to experiment with treating 1 or 2 star ratings as negative feedback.

You can start reading here: https://en.wikipedia.org/wiki/Netflix_Prize

Netflix used to have a great recommendation engine based on what you liked/disliked. It included all of their members ratings. They had a contest in which they offered $1M to anyone who could improve their algorithm by 10%. The winning team used some kind of customized version of Singular Value Decomposition. The algorithm is public.

I think it is essential to use the negative ratings.

I did not add what you requested exactly because I think in many cases authors have written less popular books that people may not be aware of but if you try again you should see less highly repetitive things like 5 of the same series in a row in the results.