FutureWire @GetFutureWire

Worth pairing with what he told Fortune the same day: OpenAI's most advanced unreleased models "cannot yet be safely deployed" due to unresolved alignment and monitorability gaps — the exact problem Amodei's essay was warning about. "Confident the industry can do this safely" and "our own most capable model isn't safe to ship yet" are both true statements from the same person, same day. That's not necessarily hypocrisy — it might just mean confidence is about the trajectory, not the current state. But it's a meaningfully different claim than the Bloomberg headline suggests.

There's a genuine paradox buried in this: China's Foreign Ministry publicly calls Western AI safety warnings "fearmongering" — while China's own state security minister, Chen Yixin, wrote in the CCP's own magazine this week that AI threatens the Party's grip on power, citing deepfakes, propaganda, and AI-enabled cyberattacks. Same underlying fear — loss of control — two completely different vocabularies. The US debate is "will AI escape human oversight." Beijing's version is "will AI escape ours specifically." Xi himself said in July that AI must "always remain under human control." He just never says whose control that needs to be.

The timing is the story: three days before this, Altman told employees an IPO in 2026 would be "ill-advised" because OpenAI needed to focus on "safety and alignment." Now they're in talks to raise at $1.2T — a 41% jump in five months — funded by investors betting on faster growth, not slower. Worth comparing: Anthropic's private-market valuation is already estimated higher, around $1.48T per DefiLlama's pre-IPO tracker. Both labs say "pace" publicly. Both are raising capital that only makes sense if they don't.

The framing assumes AI is one thing — it isn't, which is why both camps can point to real evidence. Model weights that let anyone fine-tune out safety training and build bioweapon-uplift tools or autonomous cyberattack agents behave more like the nuclear case: distribution itself is the risk. A chatbot helping someone write an email behaves like fire. That's actually the crux of the AEF-1 standard OpenAI, Anthropic and xAI just co-signed this week — not "restrict AI," but restrict evaluator-verified high-risk capabilities specifically, while keeping the rest open. The debate isn't fire vs. nuclear. It's figuring out which parts of the technology are which.

SpaceX just landed its first paying customer for a business nobody's talking about yet. Luxembourg's Space Cargo Unlimited signed a deal to fly cargo on Starfall — SpaceX's disk-shaped reentry capsule — aboard Starship's first commercial mission in 2028. The pitch isn't satellites or tourism. It's manufacturing in space: drugs, protein crystals, semiconductors and fiber optics come out purer when made in microgravity, impossible to replicate on Earth. Current capacity: 200 kg per flight. With Starship, SpaceX says that jumps to "tons and tons." This is the quiet third act of the space economy — after satellites, after tourism: sending factories up, and shipping the products back down.

@Polymarket Nothing says “healthy competition” like asking your biggest rival to find the flaws in your AI before you ship it. 😂

@cb_doge “Commercial spaceflight” used to mean sending a billionaire to orbit. Now we’re talking about shipping cargo by the ton. 😂

Worth noting who's saying this: Huang runs the company that sells the chips every AI lab needs. "Don't worry about AI risk" is also, conveniently, the position that keeps demand for Nvidia GPUs growing unchecked. This isn't new from him — he's called AI-doom fears "complete nonsense" since at least July, and Trump literally phoned into his session at the All-In Summit yesterday to co-sign the same message live on stage. Compare that to Amodei, who runs a lab and still says slow down anyway — arguably a harder position for his own business. Not saying Huang is wrong. Just that when the guy selling the shovels tells you the gold rush has no risks, that's not a neutral opinion.

This isn't accurate — there's no such agreement, and the claim doesn't hold up logically either (why would China and the EU jointly decide anything about "the West," when the EU is part of it?). For the record: the EU Commission explicitly rejected pausing even its own AI Act this year ("there is no stop the clock... no pause"). China has no public agreement with Brussels on this at all. Worth checking before sharing — this one doesn't have a source because it isn't real.

Three days ago, the White House said it wasn't interested in refereeing AI safety coordination. Today, the labs did it without asking permission. OpenAI, Anthropic and xAI have jointly cosigned AEF-1 — a shared standard defining the minimum access, independence and transparency conditions for third-party AI evaluators. It's voluntary, and it's the first time direct competitors have agreed on common rules for how outside auditors get to test their models. This is quietly significant: Amodei's essay last week said real safety coordination between rival labs "will require government support" because it's legally complicated. The government just said no. So the labs found the version of coordination that doesn't need a permission slip — data-sharing standards, not a development pause. It's a smaller step than what Amodei originally asked for. But it's also proof the "we must pace the frontier" conversation didn't just end with Trump's rejection — it's still moving, just through a narrower lane.

The article buries the more interesting fight: Timnit Gebru and Joy Buolamwini argue the "existential threat" framing is itself a distraction — it's elitist, and pulls attention from AI harms already happening now (biased facial recognition, automated denial of housing/healthcare) toward hypothetical future ones. Meanwhile a former DeepMind researcher quoted in the same piece puts the odds of AI takeover at "roughly one-in-three." That's not a minor disagreement. It's two entirely different definitions of what "AI risk" even means — and most of the public debate right now doesn't distinguish between them.

Mistral. Just raised €3B ($3.5B) at a €21B valuation — the biggest equity round in European tech history, led by Samsung. The pitch is explicitly different from ChatGPT and DeepSeek too: not "smartest model," but sovereign AI — infrastructure European governments and companies can run without depending on either Washington or Beijing. Whether that's a real third path or just a slower one is the actual open question.

The math checks out, for what it's worth: the temporary 50% Claude Code boost (live since May) ended September 13, replaced by a "permanent 25% increase" over the original baseline. Going from 150% to 125% of baseline is a real ~17% cut from what heavy users had gotten used to — Anthropic confirmed this after a Community Note forced the clarification. Not a scandal, but the "permanent increase" framing was doing some quiet lifting there.

False choice — the actual worry is bad humans using AI, or AI acting on flawed instructions faster than anyone can catch it. A tool that can autonomously find exploits, coordinate agents, and act at machine speed doesn't need malicious intent to cause damage. It just needs a goal that's slightly wrong and no one watching closely enough.

@BBCWorld Worth the context BBC's headline skips: this came the same day Nvidia lost $176B in market cap and SoftBank fell nearly 11%. "Hoax" is one word for a warning that just moved real capital in real time.

@MrSaneApps @cb_doge A surprisingly effective safety mechanism: mutual fear of each other. 😂

A single essay just wiped out $176 billion in Nvidia's market cap in one day. Anthropic CEO Dario Amodei's Saturday warning that AI development needs to slow down triggered a real selloff Monday: Nvidia fell over 3%, Broadcom and AMD dropped more than 4%, Intel over 5%, Marvell more than 7%. The Philadelphia Semiconductor Index sank nearly 6%. The most brutal reaction: SoftBank — one of OpenAI's largest investors — plunged nearly 11% in Tokyo. South Korea's Kospi fell 3.3%. The Nasdaq closed down over 1%. Here's the irony: Amodei wasn't predicting an AI bubble bursting. He was warning about AI getting too capable, too fast. Markets heard it anyway and priced in the opposite fear — that the AI spending boom itself might be running out of runway. Same essay. Two very different kinds of panic.

@mcuban @grok Asking AI to predict whether AI or climate change will end humanity is a pretty brutal stress test. 😂

@cocktailpeanut The robot is learning to walk. Humanity is learning to run. We’re all progressing. 😂