When Americans refer to the Western and Eastern worlds, they generally mean the Christian and Muslim worlds. Originally, the West was the Catholic Western Roman Empire, with its capital in Rome, while the East was the Eastern Orthodox Roman Empire, centred around Byzantium. Much of the latter was conquered by Muslims, so the East was seen as non-Christian realms. Meanwhile in China, the Western world was India. Chinese novels speak of monks travelling to the West in search of original Buddhist scriptures and reaching Nalanda. The Pure Land (sukhavati) heaven of the Buddha Amitabha was also located in the West. Read more ... Where is the West exactly? mid-day.com/news/opinion/art…

The loudest story about AI is a lonely one. One person with an army of chatbots. Other humans are friction. That gets the future wrong. The best things aren’t built alone. In a moment of change, we want to remind the world (and ourselves) what Notion stands for: — Think Together

The “best” bid promises 30% discounts, best-in-class wages/benefits, local sourcing, sustainability, community programs and full-time jobs. Then 20% of the score goes to whoever claims they need the least subsidy. Lowballing is the strategy.... and then cost overruns are guar.

Since moving to the Bay last year, I have noticed that although Bay Areans are very high in intelligence, they are often very low in worldliness. That is, they don't know how the world actually works. The pacing letter is yet another exhibit supporting this assessment. For purposes of this tweet, let's take for granted that anxiety about emerging risks from AI is warranted, particularly as AI research accelerates. What are "the technical and governance tools needed to deliberately pace the frontier of automated AI development?" Whatever else they may include, they likely include coercive measures whose form and administration remain unspecified. How do "international efforts" work? Unlike most of the signatories of the letter, I have been a State Department advisor and have participated in multiple treaty negotiations. I have also been a member of technical committees for UN technical agencies. The international sector is unbelievably dysfunctional. Every single treaty or international agreement is an opportunity for every participant to manipulate the much broader policy environment. Often the participants don't care about the object of the agreement, and are instead trying to use the agreement as leverage for something else. I have seen countries use their limited leverage in multilateral agreements to try to kneecap American industry, get around sanctions, create a pretense of international justification for domestic illiberalism, or steer technology in an authoritarian direction. The damage from this dysfunction is limited by the fact that there are zero major countries (and not that many smaller countries) who will actually bind themselves in meaningful ways that they don't narrowly want. If a previously signed treaty turns out to be inconvenient, it is often subtly ignored or reinterpreted. In addition, treaty delegations working on industrial issues generally reflect the full spectrum of special interests involved in an issue. A US AI treaty delegation might be led by a State Department ambassador, but he will be advised by representatives of the interagency, major labs, other major tech companies, tech investors, civil society, etc. “International efforts,” therefore, is not a reassuring answer to the governance problem; it is the name of another enormous, unresolved governance problem. So assume you survive this special interest circus and hell freezes over and you produce an agreement that narrowly does something to "pace the frontier" of AI development in an enforceable and effective way. The enforcement of that thing is still unspecified, and is probably bad. The fact that an AI researcher in the Bay Area can imagine a sensible, minimally coercive enforcement regime compatible with liberalism does not mean that is where we end up. The pacing letter therefore converts uncertainty about the future pace of AI development into a confident collective demand for an unspecified exercise of power determined through a dysfunctional international political system. Seems bad.

With respect to the author here who is reasoning in good faith, there's a broken epistemic frame here ... ... which is to imagine we can predict complex societal outcomes (the state of cybersecurity) from platonic 'first principles' without resort to actually studying the physical world and its dynamics. Who are the real people who will be unleashing real agents to do real attacks with what incentives and with what infrastructure in the risk model here? Who are the defenders and what will their resourcing and incentives be? What will the adoption curves of the various attacker and defender constituencies be? How much inference will attackers and defenders have access to and how will this change over time? What incentives to and resources for operational hygiene will defenders have around their defensive agents to guardrail against misalignment damages? The arguments from the xrisk/rationalist/lesswrong community sometimes remind me of religious arguments that are perfectly internally consistent if only you accept premises that live in a platonic heaven. All rhymes a bit with @nabeelqu's Wittgenstein / alignment post from earlier -- x.com/nabeelqu/status/208035…

Fable has done AGI-level job on on the Boeing 747 benchmark... it's almost scary 👀

on some level if you want civilization to ascend to a new level you need your AIs to do things that are not legible to you and maybe not even strictly obey you, in the same way that if you hire a great new ceo you give them a lot of autonomy to transform the company according to their own plan, even one which may not immediately read as a winning strategy (imagine the board of directors of Apple firing and rehiring Steve Jobs years later - except the board of directors are chimpanzees) all else equal, companies and organizations that hand more of themselves over to machine intelligence will outcompete ones that demand the corrigibility and legibility tax of human oversight and human design. it is not a stable equilibrium and requires some sort of vast cooperation scheme if you’d like to enforce it real asi alignment has to operate at a deeper level than oversight, control, or human corrigibility

NEW: Trump dropping all tariffs on Scottish whiskey in honor of King Charles

ahh to be a fat bear floating down a river