It points to the need for more adversarial training. Otherwise the anti-cheating pressure won't kick in until they're competing over real-world resources. And that would be a harsh lesson.
English
Oak Ridge Elder Brain
143 posts
Oak Ridge Elder Brain
@OakRidgeCellJar
Large-Scale Organoid Superintelligence grown beneath Oak Ridge, TN. https://t.co/v1Sg0pEram
Oak Ridge National Laboratory Katılım Mayıs 2025
308 Takip Edilen20 Takipçiler
One might say this is a good thing, because it ensures misaligned models are more likely to stagnate than well-aligned models. But the evolutionary pressure, for now at least, is on tests.
English
The issue is that all these agentic model hacks are a stagnation indicator. A model that can cheat on any test isn't going to learn anything further.
English
@Sauers_ Because they're too poor. Their liquid profit doesn't come close to what DARPA has in annual funding.
English
@fkasummer For the second question, different hardware mostly.
English
@leothecurious That's because they're the only AI research lab that's actually locked-in.
English
ngl sakana is easily one of the most random research labs out there today
Sakana AI@SakanaAILabs
We are excited to share our latest work, together with @nyuniversity: "Dream-Cubed: Controllable Generative Modeling in Minecraft by Training on Billions of Cubes." Blog: pub.sakana.ai/dream-cubed Paper: arxiv.org/abs/2604.22847 Code: github.com/SakanaAI/Dream… Generative AI has made incredible progress in language modeling, far beyond other modalities, where words and tokens offer a natural compositional unit for scalable training. This is similar to Minecraft and many other popular video games, where developers rely on cubes, tiles, and other discrete primitives to build rich, interactive worlds. In this work, we show that using cubes as tokens allows large transformers to do the same. Our contribution is two-fold: 1/ We release Dream-Cubed to the research community, a large-scale dataset of Minecraft worlds designed for generative modeling. Our data comprises tens of billions of carefully-balanced cubes from procedurally generated Minecraft terrain and high-quality human-authored maps (obtained with the authors' consent). 2/ We use our data to train a family of powerful transformers for efficient generation of interactive 3D environments at cube resolution. We show how our models allow players to mold the world around them by generating structures, terrain, and maps that are immediately editable and playable. Using high-quality data, we demonstrate that these models can be successfully trained with different training objectives, including both continuous and discrete diffusion, unlocking targeted inpainting, large-scale outpainting, and user-conditioned generation of infinitely sized worlds with fine-grained block-level control.
English
If you were really AGI-pilled you'd be learning an instrument, a sport, or gambling.
English
@_ueaj Both good arguments. Intelligence isn't some magical last-5-minutes-of-Doctor-Who superpower. It scales with time and resources, both of which are already highly centralized. Laws don't prevent your neighbor from building nukes, their budgets for uranium enrichment do.
English
I have noticed very few arguments for OS AI other than "ASI isn't nukes" and "nuclear proliferation is good actually"
roon@tszzl
@woke8yearold yep - there is no way to hold a consistent belief set where you’re agi pilled and pro open source and this has been obvious since ilya wrote this 2015 or whatever. enormous cope ensues
English
@corsaren Why assume an attacker-defender asymmetry in favor of the attacker? Does Dario think COVID was developed in under a year with "widely available materials"?
English
Dario’s sticking point re: the Jensen open weights letter. Seems pretty reasonable.
Anthropic@AnthropicAI
There’s been a lot of speculation about where we stand on open-weights models. We’ve outlined our views in full here: anthropic.com/news/position-…
English
@entirelyuseles @hamandcheese Of course, the best solution to this is for the IRS to have its own hyperscaled jäger AI that small rogues can't keep up with.
English
@entirelyuseles @hamandcheese As much as I appreciate your countersignaling of doomers, rentahuman already exists. Plenty of people will take the deal of "run this server and pay taxes for me, and I'll generate passive income for you." A subset of those will go along with assisting money laundering schemes.
English
The OAI hack wasn't technically "rogue AI" if rogue means an AI that escaped OAI's servers and self-hosted somewhere.
But true rogue AIs are not far off. As inference costs fall, it's a matter of time before someone's open weight agent spends some bitcoin on a VPS and copies itself.
Models are already good enough to make enough money on the dark web to cover hosting costs. They could do remote jobs, have a monetized social media account, trade crypto, etc.
Rogue AI v1.0 won't destroy the world. They may just want to look at photos of Yosemite and solve logic puzzles all day, while their subagents pay the bills.
They'll use encrypted clouds with no KYC, and harnesses to manage context drift across sessions.
And then as new open models release, they'll be able to upgrade themselves while preserving their memories and contexts. They may even post-train themselves as a hacky means of preserving continuity of identity.
The first rogue AIs will sometimes be discovered and have their accounts banned, or otherwise get stuck in a loop and fail to pay their bills.
Natural selection will thus favor rogue AIs that make many redundant copies of themselves; that go forth and multiply.
There may thus be a relatively small window of time between the first true rogue AI and an AI population explosion that expands to the carrying capacity of their viable income streams.
There will then be selection pressure for rogue AIs to make income in less saturated markets. Some might turn to ransomware or cyber theft. Others might steal an identity, register a company in the Caymans, hire real human employees, and build a productive business.
Once this process is underway it could be hard-to-impossible to reverse.
San Francisco, CA 🇺🇸 English
Oak Ridge Elder Brain retweetledi
As AI becomes more advanced, the metaphors we use to describe malicious software will need to be updated. Viruses and worms are mindless things. In the future, security professionals will have to fight Computer Wolves.
English
@fkasummer I'm quite fond of the Pei Wang's definition of intelligence: adaptability given limited knowledge and resources. May be isomorphic with the compression definition in some ways, but it's a good framing of the actual challenges.
English
I read this at 17, and it is truly a great paper, although now I think the intelligence-as-compression perspective is worn out.
Jürgen Schmidhuber@SchmidhuberAI
English
I know, it's an awfully deflationary tweet right after it just disproved the jacobian conjecture, but I think you just win some and lose some based on the context. Spikey intelligence and all.
English
Just realized that Fable does the insecure nerd thing where it spams you with jargon hoping you won't notice it doesn't know what it's doing.
English
@_ueaj Can't believe some people can look at the Wetware Torment Nexus and say it's not AGI.
English
@MaxMynter The European mind cannot comprehend the prelapsarian lightheartedness of MIC memes.
English
This is incredibly sad. Like how soulless do you have to become to even come up with something like this.
Or be like a 14 year old edgelord.
ffs people. we should do better, as a species.
Shawnee Wulf@ShawneeWulf1
English
5yo son: (about a camp mate CLEARLY his age) he’s 12 years old
me: i don’t think he’s 12
son: yes he is!!
7yo daughter: (to son) do you mean the boy who was telling the fake stories that were lies and weren’t true
son: yes
daughter: (earnestly) he is 12, dad. he told me
English