AI & LLMs General

We articulate a vision of artificial intelligence (AI) as normal technology. To view AI as normal is not to understate its impact—even transformative, general-purpose technologies such as electricity and the internet are “normal” in our conception

The normal technology frame is about the relationship between technology and society. It rejects technological determinism, especially the notion of AI itself as an agent in determining its future. It is guided by lessons from past technological revolutions, such as the slow and uncertain nature of technology adoption and diffusion. It also emphasizes continuity between the past and the future trajectory of AI in terms of societal impact and the role of institutions in shaping this trajectory.

While their specific features vary, a common thread between these researcher-focused apps is that they emphasize traceability of citations back to the source, even for AI-generated output. This is obviously well-aligned with the academic mindset. We expect detailed traceability will become very widespread in research-focused AI products, and may even become more prominent in general-purpose tools like ChatGPT.

As everyone who’s been through a Ph.D. knows, true insight comes from synthesizing ideas across papers and research fields. Not just from asking questions of a single source. We see more value in the growing long-context capabilities of AI models like Google’s Gemini, which can now reference dozens of full-text papers in a single chat session. We’re also excited by apps that enable accurate AI-powered searches across a wide body of sources, such as NotebookLM, Elicit and Litmaps.

How researchers use AI today

Summarization is one of the more compelling applications of AI in academic literature, so it makes sense that it was the most highly-valued among researchers.

The accusation of genocide being carried out against white farmers is either a horrible moral stain or shameless alarmist disinformation, depending on whom you ask,

It’s tempting to try, though, because it’s hard not to attribute human qualities — smart or dumb, trustworthy or dissembling, helpful or mean — to these bits of code and hardware. Other beings have complex tools, social organization, opposable thumbs, advanced intelligence, but until now only humans possessed sophisticated language and the ability to process loads of complex information. A.I. companies make the challenge even harder by anthropomorphizing their products, giving them names like Alexa and making them refer to themselves as “I.” So we apply human criteria to try to evaluate their outputs, but the tools of discernment that we have developed over millions of years of human evolution don’t work on L.L.M.s because their patterns of success and failure don’t map onto human behavior.

There’s little point in telling people not to use these tools. Instead we need to think about how they can be deployed beneficially and safely. The first step is seeing them for what they are.

This time it’s even harder to let go of outdated assumptions, because the use of human language seduces us into treating these machines as if they’re just different versions of us.

I arrived at them as a journalist who has spent a lot of time talking to the engineers building powerful A.I. systems, the investors funding it and the researchers studying its effects. And I’ve come to believe that what’s happening in A.I. right now is bigger than most people understand.

The most disorienting thing about today’s A.I. industry is that the people closest to the technology — the employees and executives of the leading A.I. labs — tend to be the most worried about how fast it’s improving.

I’ve also found many uses for A.I. tools in my work. I don’t use A.I. to write my columns, but I use it for lots of other things — preparing for interviews, summarizing research papers, building personalized apps to help me with administrative tasks.

Most of the advice I’ve heard for how institutions should prepare for A.G.I. boils down to things we should be doing anyway: modernizing our energy infrastructure, hardening our cybersecurity defenses, speeding up the approval pipeline for A.I.-designed drugs, writing regulations to prevent the most serious A.I. harms, teaching A.I. literacy in schools and prioritizing social and emotional development over soon-to-be-obsolete technical skills. These are all sensible ideas, with or without A.G.I.

This report focus on the intersection of Foreign Information Manipulation and Interference and Large Language Models. The aim is to give a non-technicalcomprehensive understanding of how weaknesses in the language models can be used for creating malicious content to be used in FIMI.

The hard reality number one is that climate change poses a severe and maybe existential threat to American security, and we haven't been taking it that seriously in the foreign policy establishment.

Climate change poses as much of a risk by the end of the century to America and its existence as nuclear warfare

AI is growing from such a small base. Beyond 2030, if AI really does take off, AI power demand globally could just begin to take over the grid. And that's already started to happen in some of the pioneer markets where data centers are particularly heavy, such as in Ireland and increasingly in the United States.

As of March 2024, there are more than 11,000 data centers worldwide, 45 percent of which are housed in the United States -

today in the United States, data centers, overall as a category, consume about 4 percent of American power

today, if we're at 4 percent in the United States for data centers overall, by the end of the decade, 2030—just five and a half years away—we are looking at anywhere from 10 percent to 12 percent of the American grid

ChatGPT query can use 100 times or more energy per query than the simple Google searc

An AI model from the University of Leeds has been trained to measure changes in icebergs 10,000 times faster than humans can.

We need to do that in order to compete with China. China is building AI data centers also at a breakneck pace, and they are not limited by the available power resources. They're actually limited by the availability of high-end chips, which America has some export controls over and is considering even more export controls, for example, on NVIDIA GPUs.

in the part of the United States with the largest concentration of data centers, right here in Northern Virginia, the power utility locally, Dominion Energy, says that it'll take more than seven years to connect the next data center of 100 megawatts or more to the grid. That's an intolerable time horizon for cloud service providers or model developers, OpenAI or lesser-known companies that are developing next-generation models.

As a result, China comes in with some inbuilt advantages in the AI race. It can power as many data centers as it needs to. In order to compete with China, the United States needs to find new sources of power, invest in our energy infrastructure, and make it possible to connect data centers much faster and have them powered with affordable and reliable energy.

Due to the predicted rise of AI data centers, utility companies have nearly doubled their forecasts on how much additional power the United States will need by 2030.

Maybe AI prose is just Prosak, the last stage of dehumanizing dullness before a revolution of human-centered, unpredictable creativity sweeps away the prose dullness we’ve been living in and reveals the genuine us: people who don’t perfectly navigate the challenges of adverbs. It’s Pat Boone singing “Tuttie Fruittie” in 1956. It’s Bing Crosby crooning “Little Drummer Boy” with David Bowie in 1977. The creators of artificial intelligence want to mimic and then bind and flatten our creativity in order to monetize it—and us. So maybe the answer is to value what’s real and original in ourselves, even if it’s not pretty.

One conclusion drawn from the study is that it raises questions about creativity, but for me, it highlights an open problem in schools about how we teach students to value depth and complexity in art and other subject areas.

We have these pieces which let us guess at the shape of the AI in front of us. It isn’t science fiction to assume that AIs will soon talk to you, see you, know about you, do research for you, create images for you - because all of that is already built, and working. I can already pull all of these elements together myself with just a little effort. That means AI can quite easily serve as personal assistant, intern, and companion - answering emails, giving advice, paying attention to the world around you — in a way that makes the Siris and Alexas of the world look prehistoric.