High-Level Machine Intelligence’

six tasks expected to take longer than ten years were: “After spending time in a virtual world, output the differentialequations governing that world in symbolic form” (12 years), “Physically install the electrical wiring in a new home”(17 years), “Research and write” (19 years) or “Replicate” (12 years) “a high-quality ML paper,” “Prove mathematicaltheorems that are publishable in top mathematics journals today” (22 years), and solving “long-standing unsolvedproblems in mathematics” such as a Millennium Prize problem (27 years)

2,778 AI researchers who had published peer-reviewed research in the prior year in six topAI venues (NeurIPS, ICML, ICLR, AAAI, IJCAI, JMLR). This to our knowledge constitutes the largest survey of AIresearchers to date

hey are experts in AI research, not AI forecasting and might thus lack generic forecasting skills andexperience, or expertise in non-technical factors that influence the trajectory of AI.

lack of apparent consensus among AI experts on the future of AI [

was disagreement about whether faster or slower AI progress would be better for thefuture of humanity.

substantial” or “extreme” concern is warranted about six different AI-relatedscenarios, including spread of false information, authoritarian population control, and worsenedinequality.

While68.3% thought good outcomes from superhuman AI are more likely than bad, of these net optimists48% gave at least a 5% chance of extremely bad outcomes such as human extinction, and 59% of netpessimists gave 5% or more to extremelygoodoutcomes. Between 37.8% and 51.4% of respondentsgave at least a 10% chance to advanced AI leading to outcomes as bad as human extinction.

the chance of all humanoccupations becoming fully automatable was forecast to reach 10% by 2037, and 50% as late as 2116(compared to 2164 in the 2022 survey).

Virtually all the policies that EAs and their allies are pushing — new reporting rules for advanced AI models, licensing requirements for AI firms, restrictions on open-source models, crackdowns on the mixing of AI with biotechnology or even a complete “pause” on “giant” AI experiments — are in furtherance of that goal.

Andrew Ng: ‘Do we think the world is better off with more or less intelligence?’

promote research, development and innovation in various data-based areas, including Big Data Analytics, Artificial Intelligence, Quantum Computing, and Blockchain.

Average is Over

that chess could be crunched by brute force once hardware got fast enough, databases got big enough, algorithms got smart enough.

Lynn Margulis [14] has made strong arguments for the view that mutualism is the great driving force in evolution.

The table below illustrates the complexity of models and data used to train common language models.

Generative AI turned one in November 2023

Sunak said that was because up to that point the government’s scientists were not pushing for it. The aim had been to “flatten the curve” and manage the spread, rather than suppress it.

AI Act: EU Commission attempts to revive tiered approach shifting to General Purpose AI

Essentially what this means is, you can test a smaller model and accurately predict how a model 106 x larger will perform.

the only differentiating factor between any two LLMs is the dataset.

to create a model with the ability to solve math problems without having previously seen them.

certain criteria

On the banks of Hallstätter See and surrounded by soaring Alpine peaks, the town of Hallstatt and its stunning landscape enjoy UNESCO protection.

Given its dual-use nature, CAI could make it easier to train pernicious systems

AI alignment, which incentivizes developers to adopt this method

CAI increases model transparency.

dependence on a specific AI technology will diminish, so that end-users can avoid ‘lock-in’ effects and benefit from reduced switching costs

Interoperability of pre-trained models across platforms should also drastically reduce the need to retrain large models.

rompt templates and standardized prompt optimizers

have assembled to develop an LLM called BLOOM10, should be valuable.

public institutions can actively incentivize data-sharing partnerships, which, in combination with federated learning, may promote AI across institutional boundaries while ensuring data privacy.

under a trustworthy and responsible governance model.

the development of a LLM is estimated to cost between 300 and 400 million euros.

source codes for formalizing the training task

AI is programmed to learn to perform a task.

arbitrarily decide

the concentration of power over technology is known to hamper future innovation, fair competition, scientific progress, and hence human welfare and development at large

concentrated power

An example is OpenAI, which was founded to make scientific research openly available but which eventually restricted access to research findings.

The Departmentwill develop and implement data qualityassessment tools and monitoring processeswith results that are transparent to users.Assessments will also be done to evaluate dataoutputs from other AI platforms to minimize risk.

High quality datasets are those sufficiently freeof incomplete, inconsistent, or incorrect data,while also being well documented, organized,and secure

he Department’s CDAO will support and coordinate the establishment and maintenance of AI policies—such as 20 FAM 201.1—that provide clear guidelines for responsible AI use, steward AI models, and prioritize the evaluation and management of algorithmic risk (e.g., risks arising from using algorithms)in AI applications during their entire lifecycle— including those related to records retention, privacy, cybersecurity, and safety.

Much like the EDS aims to cultivate a data culture, the Department will imbue its values around responsible AI use across the organization, including to uphold data and scientific integrity.

enhance AI literacy, encourage and educate on responsible AI use, and ensure that users can adequately mitigate some risks associated with AI tools.

To meet the computational demands of AI development, our infrastructure will leverage Department cloud-based solutions and scalable infrastructure services.

Robust access controls and authentication mechanisms aligned to Zero Trust principles will mitigate risk of unauthorized access to AI technologies and Department data, providing a high level of security

with a mix of open-source, commercially available, and custom-built AI systems.

ts Enterprise Data Strategy (EDS) in September 2021

Innovate

Ensure AI is Applied Responsibly

Foster a Culture that Embraces AI Technology

Leverage Secure AI Infrastructure

50 Global Hubs for Top AI Talent

if Senator Pascale Gruny has anything to say about it. She has just taken a first step toward a proposed law making everything, or really anyone — at least in official documents — well, masculine.

At Diplo, the organisation I lead, we’ve been successfully experimenting with an approach that integrates data labelling into our daily operations,

However, in an environment where data moves across borders in seconds, and can easily be destroyed or removed, cooperation based on a traditional MLAT is blamed for being slow and insufficient since it often takes months to address requests for assistance (Kent, 2015).  

US Secretary of State Antony Blinken has been touring the region again. He's currently in talks in Ankara, Turkey. And we are also being told that the head of the CIA, William Burns, who used to be the top US diplomat on the Middle East, is in the region too.

Nishida’s philosophy is critical of dualistic perspectives that often influence our understanding of humans versus machines. He would likely argue that humans and machines are fundamentally interlinked. In this interconnected arena, beyond traditional dualistic frameworks (AI vs humans, good vs bad), we should formulate new approaches to AI.

OpenAI, Google and a ‘digital anthropologist’: the UN forms a high-level board to explore AI governance

The fellows funded by Open Philanthropy, which is financed primarily by billionaire Facebook co-founder and Asana CEO Dustin Moskovitz and his wife Cari Tuna, are already involved in negotiations that will shape Capitol Hill’s accelerating plans to regulate AI. And they’re closely tied to a powerful influence network that’s pushing Washington to focus on the technology’s long-term risks — a focus critics fear will divert Congress from more immediate rules that would tie the hands of tech firms.

William Isaac

Arvind Narayanan

Sara Hooker,

Yoshua Bengio

John McDermid

Alexander Babuta

Irene Solaiman

Paul Christiano

Capabilities and risks from frontier AI

On Wednesday, the U.K. government released a report called “Capabilities and risks from frontier AI,” in which it explains frontier AI as “highly capable general-purpose AI models that can perform a wide variety of tasks and match or exceed the capabilities present in today’s most advanced models.”

that the criteria apply,

that maintains horizontal exemption conditions and largely overlooks the negative legal opinion.

harshly criticised by the European Parliament’s legal office

under a pre-set list of critical use cases were deemed automatically high-risk

I usually always thought of hackers

Inclusive Framework on BEPS

The GloBE rules provide for a co-ordinated system of taxation intended to ensure large MNE groups pay this minimum level of tax on income arising in each of the jurisdictions in which they operate. The rules create a “top-up tax” to be applied on profits in any jurisdiction whenever the effective tax rate, determined on a jurisdictional basis, is below the minimum 15% rate.

Can I capture my vision in a page? A paragraph? A word?

The European Cyber Resilience Act (CRA)

We emphasize that EU policymakers should consider strengthening deployment requirements for entities that bring foundation models to market to ensure there is sufficient accountability across the digital supply chain.

Open releases generally achieve strong scores on resource disclosure requirements (both data and compute), with EleutherAI receiving 19/20 for these categories.

the European Parliament adopted a draft of the Act by a vote of 499 in favor, 28 against, and 93 abstentions.

We are at the same juncture now with AI. Once again, a handful of competitive but ideologically aligned leaders are telling us that large-scale, general-purpose AI implementations are the only way forward. In doing so, they disregard the dangerous level of complexity and the undue level of control and financial return to be granted to them.

What is happening is not just a battle for market control. A small number of tech titans are busy designing our collective future, presenting their societal vision, and specific beliefs about our humanity, as theonly possible path. Hiding behind an illusion of natural market forces, they are harnessing their wealth and influence to shape not just productization and implementation of AI technology, but also the research.

sraeli forces were still bombing Gaza and fighting with Hamas gunmen in parts of southern Israel in the early hours of Sunday and a spokesman for the military said the situation in the country was not totally under control.

Large-scale compute is also environmentally unsustainable: chips are highly toxic to produce17 and require an enormous amount of energy to manufacture:18 for example, TSMC on its own accounts for 4.8 percent of Taiwan’s national energy consumption, more than the entire capital city of Taipei.19 Running data centers is likewise environmentally very costly: estimates equate every prompt run on ChatGPT to the equivalent of pouring out an entire bottle of water.20

access to compute—along with data and skilled labor—is a key component

Artificial intelligence (AI) is in the news every day and corporate strategies are evolving to adapt our businesses to AI use.

2 Senators Propose Bipartisan Framework for A.I. Laws

Inside effective altruism, where the far future counts a lot more than the present

The Reluctant Prophet of Effective Altruism

ChatGPT-maker OpenAI and The Associated Press said Thursday that they’ve made a deal for the artificial intelligence company to license AP’s archive of news stories.

“It’s like asking, ‘should the newsroom use the Internet?’ in the 1990s,” Tofel said. “The answer is yes, but not stupidly.”

Restore the Index by Importing the Mapping

Korisnicima treba

New Delhi’s decision reflected its “growing concern at the interference of Canadian diplomats in our internal matters and their involvement in anti-India activities”, the ministry said in a statement on Tuesday.

Timeline of Systematic Data and the Development of Computable Knowledge

I think there is fatigue. If you ask the average New Yorker what the SDGs are, I’m not sure they’re going to be able to respond.

the hardening divides between the West vs. the global South, with the two camps mainly feuding over the reform of such financial institutions as the World Bank and the International Monetary Fund.

Only 15 percent of all 17 goals have been met, and 48 percent are off track and have either stagnated or regressed.

he 80th Session of the United Nations General Assembly, a High-Level Event on Science, Technology and Innovation for Development

the role of multi-stakeholder partnerships to foster strategic long-term investment in supporting the development of science, technology and innovation in developing countries,

including the Global Digital Compact, aligned with the Sustainable Development Goals, should be considered, which should offer preferential access for developing countries to relevant advanced technologies

triangular cooperation projects

inequalities in data generation

We acknowledge that all technological barriers, inter alia, as reported by the IPCC, limit adaptation to climate change and the implementation of the National Determined Contributions (NDCs) of developing countries

We note the central role of Governments, with the active contribution from stakeholders from the private sector, civil society, academia and research institutions, in creating and supporting an enabling environment at all levels, including enabling regulatory and governance frameworks

the expansion of open-science models

the knowledge produced by research and innovation activities can have in designing better public policies

the Tunis Agenda and the Geneva Declaration of Principles and plan of action shall lay down the guiding principles for digital cooperation.

to ensure that the World Summit on the Information Society (WSIS+20) General Review process, the Global Digital Compact and the Summit of the Future contribute to, inter alia, the achievement of sustainable development and closing the digital divide between developed and developing countries.

close alignment between the World Summit on the Information Society process and the 2030 Agenda for Sustainable Development,

the full implementation of the 2030 Agenda and the Addis Ababa Action Agenda

has the potential to resolve and minimize trade-offs among the Goals and targets,

an open, fair, inclusive and non-discriminatory environment for scientific and technological development.

stakeholders

The bot is wonderful, he said, letting him speed through $10 tasks in a matter of minutes. When we spoke, he was having it rate another chatbot’s responses according to seven different criteria, one AI training the other.

here were listings for AI trainers with expertise in health coaching, human resources, finance, economics, data science, programming, computer science, chemistry, biology, accounting, taxes, nutrition, physics, travel, K-12 education, sports journalism, and self-help. You can make $45 an hour teaching robots law or make $25 an hour teaching them poetry.

One way the AI industry differs from manufacturers of phones and cars is in its fluidity. The work is constantly changing, constantly getting automated away and replaced with new needs for new types of data. It’s an assembly line but one that can be endlessly and instantly reconfigured, moving to wherever there is the right combination of skills, bandwidth, and wages.

This debate spilled into the open earlier this year, when Scale’s CEO, Wang, tweeted that he predicted AI labs will soon be spending as many billions of dollars on human data as they do on computing power; OpenAI’s CEO, Sam Altman, responded that data needs will decrease as AI improves.

Taskup.ai, DataAnnotation.tech, and Gethybrid.io all appear to be owned by the same company: Surge AI. Its CEO, Edwin Chen, would neither confirm nor deny the connection, but he was willing to talk about his company and how he sees annotation evolving.

Often their work involved training chatbots, though with higher-quality expectations and more specialized purposes than other sites they had worked for.

“If there was one thing I could change, I would just like to have more information about what happens on the other end,” he said. “We only know as much as we need to know to get work done, but if I could know more, then maybe I could get more established and perhaps pursue this as a career.”

One engineer told me about buying examples of Socratic dialogues for up to $300 a pop. Another told me about paying $15 for a “darkly funny limerick about a goldfish.”

But if you want to train a model to do legal research, you need someone with training in law, and this gets expensive.

to be looking at their accuracy, helpfulness, and harmlessness

The model is still a text-prediction machine mimicking patterns in human writing, but now its training corpus has been supplemented with bespoke examples, and the model has been weighted to favor them

Each time Anna prompts Sparrow, it delivers two responses and she picks the best one, thereby creating something called “human-feedback data.”

“I remember that someone posted that we will be remembered in the future,” he said. “And somebody else replied, ‘We are being treated worse than foot soldiers. We will be remembered nowhere in the future.’ I remember that very well. Nobody will recognize the work we did or the effort we put in.”

Training a large model requires an enormous amount of annotation followed by more iterative updates, and engineers want it all as fast as possible so they can hit their target launch date.

According to workers I spoke with and job listings, U.S.-based Remotasks annotators generally earn between $10 and $25 per hour, though some subject-matter experts can make more. By the beginning of this year, pay for the Kenyan annotators I spoke with had dropped to between $1 and $3 per hour.

nstruction writers must come up with rules that will get humans to categorize the world with perfect consistency. To do so, they often create categories no human would use.

When AI comes for your job, you may not lose it, but it might become more alien, more isolating, more tedious.

coherent processes broken into tasks and arrayed along assembly lines with some steps done by machines and some by humans but none resembling what came before.

“AI doesn’t replace work,” he said. “But it does change how work is organized.”

A recent Google Research paper gave an order-of-magnitude figure of “millions” with the potential to become “billions.”

Annotation is big business. Scale, founded in 2016 by then-19-year-old Alexandr Wang, was valued in 2021 at $7.3 billion, making him what Forbes called “the youngest self-made billionaire,” though the magazine noted in a recent profile that his stake has fallen on secondary markets since then.

Mechanical Turk and Clickworker

CloudFactory,

Human intelligence is the basis of artificial intelligence, and we need to be valuing these as real jobs in the AI economy that are going to be here for a while.”

The more AI systems are put out into the world to dispense legal advice and medical help, the more edge cases they will encounter and the more humans will be needed to sort them.

Machine-learning systems are what researchers call “brittle,” prone to fail when encountering something that isn’t well represented in their training data.

The resulting annotated dataset, called ImageNet, enabled breakthroughs in machine learning that revitalized the field and ushered in a decade of progress.

The anthropologist David Graeber defines “bullshit jobs” as employment without meaning or purpose, work that should be automated but for reasons of bureaucracy or status or inertia is not.

But behind even the most impressive AI system are people — huge numbers of people labeling data to train it and clarifying data when it gets confused.

Like most of the annotators I spoke with, Joe was unaware until I told him that Remotasks is the worker-facing subsidiary of a company called Scale AI, a multibillion-dollar Silicon Valley data vendor that counts OpenAI and the U.S. military among its customers

oe got a job as an annotator — the tedious work of processing the raw information used to train artificial intelligence.

LLMs are through and through based on language and patterns to be found through it.

the very success of LLMs in the commonsense arena strongly suggests that you don’t fundamentally need deep “structured logic” for that.

One of the surprises of LLMs is that they often seem, in effect, to use logic, even though there’s nothing in their setup that explicitly involves logic.

“symbolic discourse language”

particular formal system that described certain kinds of things

we weren’t trying to use just logic to represent the world, we were using the full power and richness of computation.

Lenat–Haase representation-languages