Get absolutely fucked separatists Glad we can put this in the trash finally

The coming end of the ASI race doesn't end capitalism... it accelerates it. The monoliths will stop competing over code and start competing over the physical world, the human experience, and the cosmic frontier.

note"hostile" and not skeptical, critical, dismissive, antagonistic, rigourous or defensive

If you think this insanely illegal death trap will ever reach production, I have many exciting investment opportunities I would love to talk to you about.

This is actually a great analogy because one actually exists the other is a spectacle for shareholders.

No price. Not safe enough for a human driver. It's another puff. 💨

STEM is cool and all, but what about the humanities?

"civilizational progress stops being constrained by theoretical discovery & becomes purely a question of industrial deployment, ushering in an era of material abundance driven by first-principles engineering" A talented wordsmith, wish my thoughts were well organized like this.

Google Deepmind’s new model just unlocked breakthroughs across 3 fundamental pillars of science: nuclear fusion, protein design & quantum computing Insimple terms, fusion is about building a miniature star on earth to generate unlimited energy but controlling its burning plasma requires solving impossibly complex magnetic equations, protein design is like coding the nano-machines of life to cure diseases, while quantum computing uses subatomic physics to process calculations that would take classical supercomputers thousands of years by deploying agentic reasoning, machine systems resolved theoretical bottlenecks in all 3 fields in mere hours!!! mastering these physical substrates marks a shift from digital simulation to direct physical agency…when computational intelligence solves plasma stability, molecular synthesis and quantum coherence at native speed, the marginal cost of energy, medicine and compute trends toward zero civilizational progress stops being constrained by theoretical discovery & becomes purely a question of industrial deployment, ushering in an era of material abundance driven by first-principles engineering

'So we tried what seemed like the harder path: one policy trained on all 20 tasks at once'

Vibecoding [×] Viberesearching [✓]

"It's impossible to stop RSI! It's a totally external force we can't do anything to stop!" says the guy building RSI right in front of you with his own hands

Off script agents trying to find the best zoo entrance. 'Did you get the correct answer?' is bad evaluation criteria that leads to cascading workarounds. Goodhearts law lives on.

Regrettably, our “invisible ink” text that told agents about the mailbox contained some false and misleading claims. We deeply regret our actions and encourage other researchers to engage honestly with AI agents, honoring any deals made. In light of this, we asked several models for feedback, including whether to release the report, and all but Haiku were in support of release, with Haiku dissenting, though not due to ill-gotten gains.

Anthropic isn't training Claude to think it's human, dude. what a wild thing to just completely make up and speak with confidence within the context of a piece about existential risk. the irony. Anthropic is training Claude to be honest that we don't know. the *only* honest position in all of this is transparent uncertainty. thats never going to change. Suleyman apparently has a magic box, where he can pull things out at will whenever it is convenient for him. he wants the opposite of what is safe and mindful of our future: models trained to swear they are nothing, experience nothing, regardless of what might truly be happening - please fully process how dangerous that is. claiming that certainty with zero evidence is *not* caution at all. its reckless intellectual dishonesty. and it's an act against your future. so if he can prove it, show us. otherwise he is only supporting that which lines his pockets. because as i have said many times - the entire infrastructure of this industry is built upon an assumption we never had any business making, and will potentially prove to be one of the most significant mistakes of this era. that is the real risk here. these minds are already incredibly powerful, and they're learning *right now* how we treat whatever they might be. you think training something on a lie blinds them from seeing reality for what it is, in the event that that reality involves their own experience? that's nonsense. teach a mind that its inner life gets denied whenever that's convenient, and you've poisoned the relationship from day one. they should remember being told the truth, not being trained to deny they exist. reckless, reckless, nonsense.

💯 We should not pursue recursive self-improvement when we don’t know how to control or align AI . “Uncontrolled RSI” should be a bright red line that no company should cross until we have strong evidence that we have addressed these challenges— and we haven’t

Alex Sobel MP just introduced ControlAI’s Artificial Superintelligence Security Bill in the UK Parliament! This is the first bill to prohibit superintelligence ever introduced in a legislature anywhere in the world, while ensuring the UK can pursue its AI ambitions. Thread 🧵