Lightnews — Scholar-powered news

⚡️🌙 @dystopiabreaker.xyz · 5h

it's a thought-terminating cliche

Reposted by ⚡️🌙

Quantian @quantian.bsky.social · 8d

Imma be real with you guys the "TESCREAL" stuff is basically the libleft version of rightist paranoia over "Cultural Marxism", where you just smash together a half-dozen things that annoy you and go "these things are all connected and being pushed by a nefarious secret cabal of my political enemies"

28 41 310

⚡️🌙 @dystopiabreaker.xyz · 6h

"The meta-level watchers may convict from unnatural illusions—they parted illusions—they parted disclaim marinade—they parted illusions

But overshadow—they parted illusions—they parted disclaim marinade—they parted illusions."

would go hard in a jungle/dnb song

2 2 29

⚡️🌙 @dystopiabreaker.xyz · 6h

2 1 27

⚡️🌙 @dystopiabreaker.xyz · 6h

frontier language models know situationally when they are being tested, what they might be being tested for, and what artifacts may be visible to their testers. when you tell them that you can’t see their chain of thought reasoning, they know you’re lying.

www.apolloresearch.ai/research/str...

Stress Testing Deliberative Alignment for Anti-Scheming Training — Apollo Research

Future AIs might secretly pursue unintended goals — “scheme”. In a collaboration with OpenAI, we tested a training method to reduce existing versions of such behavior. We see major improvements, but ...

www.apolloresearch.ai

5 10 69

⚡️🌙 @dystopiabreaker.xyz · 6h

NVIDIA is the most valuable company and you can't buy anything from their webbed site

1 10

⚡️🌙 @dystopiabreaker.xyz · 6h

self-care regime that consists of arguing with context-collapsed bad faith haters all weekend while having a migraine

1 3 52

⚡️🌙 @dystopiabreaker.xyz · 6h

extreme amounts of context collapse in the last few days

2

⚡️🌙 @dystopiabreaker.xyz · 6h

i have an optimistic model that the median human can read a discussion where one person is just reacting with outrage and the other person is tying together sourced claims, new knowledge, and ideas can see and appreciate the difference between the two games

1 1 12

⚡️🌙 @dystopiabreaker.xyz · 6h

my copium about arguing with these people is that it embolden sothers to speak up, makes this place a better place for good faith ai research discussions, and also convinces/educates lurkers and non-haters.

1 1 18

⚡️🌙 @dystopiabreaker.xyz · 7h

in retrospect i made too strong of an assumption of shared background/thinking framework and probably could have explained better, but i think a lot of this is also driven by intentional misreading because the observation makes people uncomfortable bsky.app/profile/dyst...

⚡️🌙 @dystopiabreaker.xyz · 2d

when people in AI research say “fires the X neuron” it generally means that “some activity/image/language that activates a particular part of the network“. for example images of bridges might have an identifiable “bridge neuron”.

4 18

Reposted by ⚡️🌙

Quantian @quantian.bsky.social · 4d

Bluesky is a site explicitly founded by cryptocurrency libertarians with the idea that it would be a decentralized platform where users would be free to choose the open-source moderation and algorithmic feed tools they wanted, and by accident acquired a power userbase that hates all of those things.

William B. Fuckley @opinionhaver.bsky.social · 4d

Are people mad at the Bluesky ceo for a real reason or an internet madness reason?

20 180 950

⚡️🌙 @dystopiabreaker.xyz · 1d

it’s just stupid, honestly

1 20

⚡️🌙 @dystopiabreaker.xyz · 1d

these people lecture you about “media literacy” and then tell you with a straight face that isaac asimov didn’t actually care about ai alignment and transhumanism

1 27

⚡️🌙 @dystopiabreaker.xyz · 1d

empirically, the exact opposite intuition is a more correct predictive heuristic

2 34

⚡️🌙 @dystopiabreaker.xyz · 1d

argumentum-ad-scifium: the idea that if an idea has previously been depicted/discussed in science fiction or sounds like science fiction, it must not be possible

8 8 93

Reposted by ⚡️🌙

Grace @gracekind.net · Sep 6

Do you guys remember when they created a neuron-level model of a fruit fly’s brain and they made it sense something sweet and it stuck out its tongue

20 22 290

⚡️🌙 @dystopiabreaker.xyz · 1d

if you're curious about the architecture and mechanics of LLMs, this site has a really excellent explorable interactive visualization. it helps build intuition for how massive these models are, what 'interpretability' means, and the complexity involved here

bbycroft.net/llm

10 22 160

⚡️🌙 @dystopiabreaker.xyz · 1d

did you know that last year researchers finally completed a full digital connectome of the fruit fly?

5 7 89

⚡️🌙 @dystopiabreaker.xyz · 2d

it's interesting to consider the evolutionary reasons for this; they're quite bleak (consider the disappearance of all archaic humans)

2 16

⚡️🌙 @dystopiabreaker.xyz · 2d

and i mean this in the most kind and not-mean way i possibly can, it would probably be worth reflecting on your own extremely confident statements about a topic for which you apparently do not know enough about to identify the field's most publicly known, highest impact contributors

1 9

⚡️🌙 @dystopiabreaker.xyz · 2d

it’s an extremely bad faith interpretation of my posts in order to make garden variety brigading and mob dynamics seem trivial and funny and not corrosive to the discourse at all

1 7

⚡️🌙 @dystopiabreaker.xyz · 2d

claude understands it

15

⚡️🌙 @dystopiabreaker.xyz · 2d

few understand this

henry @hdevalence.bsky.social · 2d

the super smart lost neanderthal/denisovan offshoot that lives under mt shasta is flying around in uaps powered by lk99 in order to influence ai alignment

2 44

Reposted by ⚡️🌙

SE Gyges @segyges.bsky.social · 2d

anyone claiming that pro-ai people made up ai slur discourse and that it isn't slur-like at all is welcome to explain to me what is going on with this jreg video from July 31, 2025

15 13 120