31–40 of 212 posts
Mixed feelings about the case. Sharing is the best thing that can happen to knowledge. It is great that gatekeepers lose money over this. However, the blame of the loss might burden oclc, which might have been doing a positive job. https://www.oclc.org/en/about.html
What is OCLCs added value? They didn't create the data
Not as valuable as the actual data, but its not nothing either.
At this point, we need a service that "offers" an 8-bay (with 12TB? 14TB? drives) full with the whole ~80TB Anna's Archive. It's essentially all of human knowledge and to be frank it belongs to no one - rather...everyone. People can store this at their house, keep it offline. Just to have these seeds of knowledge everywhere. ...I suppose LLM's trained on this data, essentially their model weights and tokenization are…
You dropped a 0. Anna's Archive is currently 862.4 TB
They're just piling on anything they can regardless of reason. A real damage count would be lost sales due to web scraping, but they don't sell anything. > For example, the organization spent $1,548,693 on upgrades for its hardware infrastructure, and an additional $608,069 for a two-year Cloudflare contract [..] Other costs include the salaries of 34 full-time employees, who were tasked with mitigating the harm caus…
Ask Aaron Swartz
They're just piling on anything they can regardless of reason. A real damage count would be lost sales due to web scraping, but they don't sell anything. > For example, the organization spent $1,548,693 on upgrades for its hardware infrastructure, and an additional $608,069 for a two-year Cloudflare contract [..] Other costs include the salaries of 34 full-time employees, who were tasked with mitigating the harm caus…
This is also why having a default judgement delivered against you for failing to show up generally isn't great.
What is the deal with this Njalla hosting service? Is it really so hard to take sites from there down?
It's made by a former piratebay founder. When you buy a domain from them, they act as a middleman and keep legal ownership of the domain thus shielding your PI somewhat from legal enforcement. They claim their hosting is "In secret locations in Sweden". Their LLC is located in Nevis (some tax evasion island). There's also https://1984.hosting/ which runs a similar operation out of iceland.
I feel obligated to say my usual "IP is at this point doing more harm than good" spiel here but don't have the time budget to argue it with people today
rms suggests dialling back copyright rather than completely abolishing it: 10 years from date of publication. Of course he doesn't believe in copyright for software at all, but that's another matter.
The funny thing is the way these greedy assholes in the copyright industry are behaving is just making it worse for them. It's driving people to places like z-library because essentially everything is in copyright. A child has just been born who won't ever see a work that was published 50 years ago go out of copyright. It's insane. With sensible copyright lengths we wouldn't need z-library.
I feel obligated to say my usual "IP is at this point doing more harm than good" spiel here but don't have the time budget to argue it with people today
What would we use instead?
Academic publishers should not exist, research - especially publically funded research - should just go straight to the public domain.
"Whether we’re supporting advancements on the leading edge of science or helping children build a strong learning foundation, shared knowledge is the common thread". (source: https://www.oclc.org/en/about.html ) Well, now it's shared on a torrent, but I guess for them that was "over shared" lol. Also, they are a library but spent 5 million on cyber defense... seriously??
Earlier quoted context omitted.
If the content is to be trustworthy then using LLMs to compress it makes no sense.
It's possible to do lossless compression with LLMs, basically using the LLM as a predictor and then storing differences when the LLM would have predicted incorrectly. The incredible Fabrice Bellard actually implemented this idea: https://bellard.org/ts_zip/
Use a universal function approximator to approximate the universe, seek Erf(x)>threshold, interrogate universe for fresh data, retrain new universal approximator, ... loop previous ... , universe in a bottle.