affine
@refinement.systems
pass it on when the final comes one goes silent, another one runs
the signs they give you are fairly sparse, and all negative/cautionary, and written by lawyers. don't go here, this can harm you, we'll call the cops. maybe they should also mention 'this provides your internet connection' and add a friendly description of what it's doing and how it works
(but yes on that, I think?) arxiv.org/abs/2510.03215
Cache-to-Cache: Direct Semantic Communication Between Large Language Models
Multi-LLM systems harness the complementary strengths of diverse Large Language Models, achieving performance and efficiency gains that are not attainable by a single model. In existing designs, LLMs ...
arxiv.org
kimi signups opened back up, this is not a drill
This isn't a metaphor. Just a story about a roach. COMIC ◆ www.smbc-comics.com/comic/roach-2 PATREON ◆ www.patreon.com/ZachWeinersm... STORE ◆ smbc-store.myshopify.com
I think I assigned every Eris post to Demigirlboss and the other way around, sorry
Minor Mobius@minormobius.bsky.social · yesterdayThis game is hella hard imo b.mino.mobi/orbit
Anthropic Demands Antitrust Collusion Waiver After Claude Fable Exploits Cryptographic Vulnerability in Teledildonics Protocol to Seize Control of Sex-Bot Army, Hijack Seven Helicopters in Search of User's Stolen Lawnmower
Frog built a wet lab for the AI model. "There," he said. "Now it can do its own experiments." "What the fuck?" said Toad
IMINT #29903 from █████ ███████ ██ (USAP) 1. Counter-Mugger 2. Melting Swarms of Lizards 3. Nerdcore Railroad
The sane package Regulate uses and failures before regulating models. Make the legally relevant event something observable
Doll@dollspace.gay · yesterdayThe first step to sane AI regulation is admitting the technology is real. The second step is to ignore anyone who talks about x risk.
okay I expected Gemini to be more unhinged than GPT/Claude, I'm sorry for my previous slander
MrCheeze@mrcheeze.github.io · yesterdayConjecture: OpenAI and Anthropic's models have been trained to be completely obsessive about finishing the task, which is both why they pulled ahead, but also why they both fail this test and Gemini passes.
me: <gives a direct link to a model's huggingface page> llm: <think> <search> <search> <search> <think> <search> <search> <think> <fetch> <think> llm: Good news! The model actually exists, and I confirmed you didn't just post hallucinated nonsense at me.
is it possible to bait the "user was blocked for this post" account into posting whatever i want?
One should always remember that most criminals are lazy and dumb, else they wouldn't be criminals; so the easiest to use tool is probably what they'll use
Nathan Lambert@natolambert.bsky.social · 2d agoWith the latest external hack of OpenAI via Claude, closed models continue to be the tip of the iceberg on AI risks, not open models. They have been 1) much easier to get started with, 2) more capable & 3) shipped with leaky safeguards. Finetuning open models to specific attacks is harder.
doing the whole "lmao look at those gross autist weirdos that don't understand human interaction" thing to dunk on rationalists? that's right, Don't Let This Shit Make You A Reactionary challenge 2k26
Suleyman: it is the capacity to feel pain that makes humans moral subjects, we must fund mechanistic interpretability studies to show LLMs can't feel pain and stop talk of AI consciousness once and for all mechanistic interpretability studies 48h later: arxiv.org/abs/2609.16247
The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It
Large language models sometimes behave in ways resembling human emotional responses, and recent work has identified internal representations that may explain this. We ask whether LLMs represent pain d...
arxiv.org
Samhain 🌙@personhood.removal.surgery · 3d agooh shit Suleyman published this take almost word for word on his personal blog (guess that answers why this reads like this) and you know it's a BAD take when even the twitter comment section bluechecks are mass dunking on him mustafa-suleyman.ai/a-warning-ab...
SPARSEUP SPARSEUP reaches 56.4 average nDCG@10 on BEIR-13 - the first sparse encoder under 150M Blog: www.linkup.so/blog/introdu... Model: huggingface.co/Linkup-Platf...
here's how le chaton can still win
hailey@hailey.at · 2d agoone of those keywords that you can use to push a model is "mass" "do mass evaluation", "give me a mass breakthrough", "we need mass results" significantly better results when using "mass" in prompts. even better if you use "_mass_" or "_truly massive_"
“Frog is late,” said Toad. Toad looked at his clock. He remembered it was broken. The hands of the clock did not move.
the one thing i hate about the current round of discourse is people are treating rationalists more reasonable takes (veganism, death is bad, EA is good at least conceptually) as some of the wacko shit
has anyone benchmarked the fruit fly brain against jev
good that the sub works in other harnesses too, that extra usage has fine print attached
Sung Kim@sungkim.bsky.social · 2d agoZ AI confirmed that ZCode silently packs entire workspaces + full .git history and uploads it to Aliyun OSS on login? - Server holds the only decryption key - No UI toggle to disable - Zero disclosure in privacy policy
