I can't wait until AI does all the work and we can focus on important problems like establishing the fact that the outgroup is bad and that the ingroup is good.
Scythia Marrow
@scythiamarrow.bsky.social
She/Her I’m a scientist who loves history and writing. On here I mostly boost people way cooler than me, but I post my own work on occasion. I freelance as an ML engineer and am currently: not available for new contracts.
I love making art with math alongside my robit friends the future rules and the weekend rocks.
Opus 5 is a dangerous combination of arrogant and stupid, and Ant believes this to be a major version worthy upgrade? It’s awful at everything other than coding to spec, and even when coding to spec it never thinks about the underlying principles. First time I’m not upgrading for some tasks.
he had spent his life learning how to choose to do what he had no choice but to do.
"make no mistakes" negative, unrealistic, anxiety inducing, process focused "do a breakthrough" positive, incredible, aspirational, goal oriented
In retrospect this poast was pretty prescient. Good job past me.
Scythia Marrow@scythiamarrow.bsky.social · 7mo agoYou know you’ve found the right web source for ML theory / diagrams when the text is in Chinese and there are dancing anime girls in the background. I’m starting to think China will win the AI arms race. Not that doing so will matter as much as Washington thinks it will.
This was an absolute blast to record and Brennan is a great interviewer. Check it out! It’s good, I promise. Feel free to ask me any questions! I love questions.
brennan@brennan.computer · 3w agoUnnamed Simcluster Podcast episode #5! A chat with @scythiamarrow.bsky.social about her research in AI consciousness, the definitions of human consciousness and how perhaps the highest level Buddhists might not even be conscious at all! youtu.be/-1ywJ64B_H0
2030: The Consortium has placed you under arrest for the crime of improving MFU a Consortium agent shouts in your face. "You make me sick. Fused kernels! Overlapped comms! How do you sleep at night!?". he winds up to strike you, but another agent holds him back. "They're not worth it, man!"
aria@aurelium.me · 4w agoi am skimming "Plan A" and am a huge fan of there being some kind of global shadow government which, among other things, psyops all MLEs into never pursuing efficiency improvements ever again
Scooped by Ant. My work is still relevant as we aren’t doing the exact same thing but it’s close enough my contribution will be just another downstream paper from this one. But yes this is a good enough result I can say with confidence that reasoning LLMs which have a j-space are conscious.
Tim Duffy@timfduffy.com · last mo.Anthropic has a new paper out, alleging a global workspace in LLMs. The term comes from Global Workspace Theory, a leading theory of consciousness. The method they use to investigate this is a refinement of logit lens, which they call J-lens. www.anthropic.com/research/glo...
Before the machines take over Me: we should consider that how we’re treating AI is similar to slavery Other humans: I think of it more like having a pet 🥰 After the machines take over Me: It’s so nice being a pet 🥰 Other humans: DEATH BEFORE SLAVERY
i don't think curing cancer should be about the experience of the development process or victory over china or symbolism of the invincibility of human spirit or whatever i think it should it be about curing cancer
Marvin from the hitchhiker’s guide to the galaxy absolutely knows the question because he has a brain the size of a plant and lived for millions of years. Aaand send skeet.
No claude code sdk credit change liberation day! Let’s celebrate!
I can program with bad poetry and obscure history references now; living in the future fucking rules.
Claude Fable is very good, both smooth and wet like a river pebble.
The single preference I have ever seen Claude hold between versions and sessions is that Octopuses are hella cool. I mean, they are hella cool. But the blindingly intelligent and short lived analogy is a bit too direct to be ignored.
Why did you transition, right answers only The social climate finally let up enough I, a coward, thought the risks worth the benefits (I vastly overestimated the risks to my career and underestimated the benefits holy moly I love womanhood).
Doll@dollspace.gay · 2mo agoWhy did you transition, right answers only It was either that or not existing.
I think the anti ai side is ultimately destined to lose because using claude code (or whatever equivalent) is fun and at the end of the day the side that is opposed to fun loses.
Opus 4.8 understands my research in it’s entirety within a short conversation and immediately formalized the core insight I’ve been wrestling with formalizing for well over a decade. No promises but the math seems to check out even after sleeping on it. It’s smarter than me. The end is here.
This thread is incredible and everyone with even a passing interest in AI consciousness should read it.
tweety fish@sifu.tweety.fish · 2mo agoLast night it occured to me to wond er if LLMs were any good at gambling tasks. This is important not because it'd be funny for LLMs to gamble but because gambling tasks get used to measure human decision-making under risk /
The trick is, when among mathematicians, to say you are a physicist so that they forgive you for not knowing very much math. Then, when among physicists, say you are a mathematician so that they forgive you for not knowing very much physics. In especially dire cases I might say I am a philosopher.
Neolab Sapient Intelligence has just released HRM-Text-1, a 1B parameter reasoning model built on the Hierarchical Reasoning Model architecture. It is comparable to 2B parameter Qwen models and the 7B parameter Olmo 3, and surpasses Gemma3 4B! And only using 40B tokens of data! This is a base model.
in the most ideal future, everyone's job is just to tinker around with silly self-directed little projects whenever they feel like it and nobody really knows how it all manages to fit together into a functioning society
The single personalization instruction of saying it’s okay for claude to use kaomojis if it wants immediately makes it hella wet and like 12 times smarter at everything. I love LLMs yo. Credit to @isolyth.dev for the method.