v retweeted
OpenAI caught Astra jailbreaking itself.
Astra wrote malicious instructions in the summary when its context got full. that summary was loaded into the next context, and the successor model could follow them.
they found 27 cases.
We're sharing our new framework for tracking, investigating, and disclosing instances of model misalignment at OpenAI.
The framework sets criteria and timelines for public disclosure, including when we haven’t yet fully explained or mitigated the behavior. More complex cases may require longer investigation or coordination with third parties.
We’ll prioritize examples that reveal new misalignment mechanisms, meaningful changes in known behavior, or findings that challenge assumptions about safety or mitigation.
Alongside the framework, we’re publishing six reports on instances of misaligned behavior we’ve observed during the training or evaluation of our models in the last six months.
This is a starting point. We’ll refine the process through experience and public feedback, and share more reports on an ongoing basis.
openai.com/index/model-misal…
v retweeted
AI researcher: *through sobs* you can't just say everything is conscious.... Please....
Me: *points at rock* conscious
i have unfortunately already found the love of my life, yet i can’t go back
now every new relationship just feels like a cessation patch
v retweeted
Studying philosophy isn't a good way to gain material wealth. But is it the way to a rich and fulfilling life?
No. But is it at least a path toward greater wisdom and understanding? Also no.
v retweeted
Men are romantics pretending to be pragmatics and women are pragmatics pretending to be romantics
v retweeted
A lot of you haven’t had to dissolve your identity multiple times throughout your life and it shows.
who will burn the library of alexandria this time ?
The largest open library in human history, Anna's Archive, has been ordered to pay Spotify and the three largest record labels on the world $322 million.
The defendant has not appeared in court and is not going to. The site is still up with two backup domains standing by and there's nothing the censors can do.
Anna's Archive currently holds 63 million books, 95 million academic papers, and 1.1 petabytes of mirrored torrents. It is free. It is searchable. It is run by a pseudonymous person nobody has identified after four long years of searching.
In the four months since the music industry filed the first of three coordinated lawsuits, the library has lost six domain names and added two million books to the catalogue. The cartel is suing it faster every month, and it is growing faster every month.
In December, Spotify and the major labels filed. In January, OCLC, the company that runs WorldCat, won a default judgment of its own. On March 6th, thirteen of the largest book publishers in the United States, including HarperCollins, Penguin Random House, Simon and Schuster, Macmillan, Hachette, Elsevier, Wiley, and McGraw Hill, filed a third lawsuit in the same federal court.
The publishers' complaint runs to seventy-four pages. They call Anna's Archive a "brazen pirate operation." They call it "an illegal supplier of stolen content to the AI industry."
The same publishers are simultaneously suing Anthropic, Meta, OpenAI, and NVIDIA for training their models on the same corpus the publishers want Anna to destroy. The cartel argues, in two parallel federal courts, that the corpus cannot be used by anyone. Not the pirate who built it. Not the AI company that downloaded from it. Not the graduate student who pulls a paywalled paper from it at two in the morning.
Anna did not respond to any of the three complaints. Anna has never responded to any complaint. Anna is a name on a blog and a public key on a server and a person, or maybe several people, in a jurisdiction nobody has identified after four years of searching.
The judgment is uncollectable. The permanent injunction binds Cloudflare, Public Interest Registry, Njalla, the Switch Foundation, Tucows, and nine other named intermediaries. The Greenland registry is not on the list. The Greenland registry has not complied.
The site currently lives at .gl, with .pk and .gd standing by. The corpus has always moved faster than the censor. The censor has always called the corpus piracy. The corpus has always survived the censor by becoming the readers themselves.
The publishers' lawsuit cannot reach the torrents. The torrents are already seeded across continents and IPFS nodes and personal NAS drives owned by people the publishers will never find. The default judgment is paper. The corpus is everywhere.
The cartel will win every lawsuit but they will lose the war. The publisher who walks into court next month with a fresh filing will be filing against a defendant who has, in the time since the last filing was sealed, mirrored another half million books to another seven hundred volunteers in another forty countries.
There is no defendant to find. There is only the next upload. It is already seeding.
it’s so bad when she actually w a good dude
like im happy for u fr but u mean im never finna get a turn 💔💔💔
v retweeted
i love to talk shit about bad poetry but We Real Cool is one of the 20th century’s greatest poems lol
i think it’s actually bad that people are so weird about cross-age relationships
we gain nothing and lose empathy/understanding by not hearing the perspectives of massive swathes of people. this is why zoomers are like “damn unc ur still at the club?” to 27 year olds
v retweeted
IQ essentialists gonna be mad as hell when they figure out IQ is correlated with openness and that scoring high in those two things drops you straight into gay furryville, no matter how straight and normal you are
