Geraldine Seymour

@geraldineseymour.bsky.social

PhD Candidate | Luddite | Artificial Intelligence in Science Fiction | Space | Gothic | Cats |Writing | Bad Drawings in Ink

OpenAI and Microsoft admitted LLMs are predatory tech built on “an astonishing theft of unprecedented proportions,” and the “largest theft of labor in human history.” An internal Microsoft document said AI has created a “doom loop” that is killing “the entire web.” www.404media.co/doom-loop-op...

‘Doom Loop’: OpenAI and Microsoft Admits LLMs Are Destroying the Web and Built on Theft

"Millions of people around the world will soon consider large models ‘hoovering up’ all their work to be an astonishing theft of unprecedented proportions."

404media.co

It seems to me that "AI may be misaligned with human values and do things we can't predict or understand" and "AI is a shitty, unpredictable tool that provides wrong answers all the time" is fundamentally a difference about whether you think LLMs are conscious beings. They are not.

In a historic move, the SF Collections at the University of Liverpool have been awarded Designated Status by Arts Council England. The award recognises them as a site of national & international significance. Many congratulations to our Curator, Tom Dillon, for successfully piloting the application.

I love this. So British media in the silly season. And we’ve only just got over the annual ‘blonde girl(s) screaming over A Level results’ story headlining in all media outlets, as it has been for the last five decades.

Four in 10 men believe they could qualify for the 2028 Olympics

Athletes who compete in the Olympics have usually spent most of their lives preparing for the event. But four in 10 British men believe they could qualify for the next Games if they started training n...

mol.im

AI did not and cannot do such a thing! it is literally in the next line (Scientists trained AI on libraries of DNA and then asked the model to create recipes for viral genomes) but hype and clickbait over accurate reporting i guess www.nytimes.com/2026/08/06/s...

This A.I. Just Created Viruses Not Found in Nature

Scientists trained artificial intelligence on libraries of DNA and then asked the model to create recipes for viral genomes. Sixteen of them were viable, yielding new viruses.

nytimes.com

Unusually for bsky, this has attracted a lot of folks who want to argue (via QTs here and downthread) that the anthropomorphizing language is just fine, actually. Helps people understand clearly. Doesn't obfuscate responsibility. And/or is a more accurate portrayal of what the models are doing. >>

Emily M. Bender@emilymbender.bsky.social · 2mo ago

This is laughable --- "reporting says", well yes, these reporters did a bang-up job anthropomorphizing the system they are talking about, and obfuscating the responsibility that OpenAI holds here. bsky.app/profile/mims... Some examples:

I cant help but see this as openai trying to do the same doom slop stunt as anthropic. But they're going further with the trust us this is super serious and more important than when anthropic said all that about mythos

“our agents escaped containment” sounds a lot more impressive than “our SDET forgot to disable the internet gateway and our agent was able to connect to a live system by mistake”

Observe the rank anthropomorphization here. Models don't "conclude" anything, nor do they "hack", "wriggle", or "cheat". You simply can't write clear informative articles about this stuff if you displace agency away from the people (organized into corporations) responsible. /fin for now.

Screencap:
Rogue AIs going unsupervised for days, hacking at least one company, represent one of the first real-world examples of a threat long feared by AI safety researchers: loss of control. It comes as the cybersecurity capacities of top models, including Anthropic’s Mythos, have helped fuel debate in Washington about how to regulate artificial intelligence.