Stefaan Verhulst

Essay by John G. Palfrey: “…The world would be different if large, open datasets could be accessed at low cost by civil society actors, provided that they incorporated constraints to limit the dangerous uses of the same technologies. Recall the example of climate change, which posited that an open-source dataset, comprising various actors, methods, and geographies, could be used to identify and enact solutions to climate issues around the world in a fraction of the time it takes today….

Philanthropy can—and should—seek to help shape technologies for the good of humanity, rather than for profit. If we do not intervene in the public interest, we may find ourselves being haunted by this missed opportunity for a brighter future. Our previous approaches to investing in and governing new technologies have left too much power in the hands of too few. The harms associated with a laissez-faire approach in an era of artificial intelligence, as compared with the previous digital technologies, may be far greater. Promises by the tech industry, from the mid-1990s to today, to self-regulate and include community members in their growth and design have not come to fruition, but they can serve as a sort of reverse roadmap for how to imagine and design the next phase of technological change. We know what will happen if a laissez-faire approach predominates.

We need to learn from this past quarter-century and design a better, more public-­interested approach for the decades to come. This moment of inflection allows us to use futurism to guide today’s investments, to remind ourselves that we can embed greater equity into the technology world, and to recommit to philanthropic practices that help to build a safe, sustainable, and just world…(More)”.

Article by Cal Newport: “Much of the euphoria and dread swirling around today’s artificial-intelligence technologies can be traced back to January, 2020, when a team of researchers at OpenAI published a thirty-page report titled “Scaling Laws for Neural Language Models.” The team was led by the A.I. researcher Jared Kaplan, and included Dario Amodei, who is now the C.E.O. of Anthropic. They investigated a fairly nerdy question: What happens to the performance of language models when you increase their size and the intensity of their training?

Back then, many machine-learning experts thought that, after they had reached a certain size, language models would effectively start memorizing the answers to their training questions, which would make them less useful once deployed. But the OpenAI paper argued that these models would only get better as they grew, and indeed that such improvements might follow a power law—an aggressive curve that resembles a hockey stick. The implication: if you keep building larger language models, and you train them on larger data sets, they’ll start to get shockingly good. A few months after the paper, OpenAI seemed to validate the scaling law by releasing GPT-3, which was ten times larger—and leaps and bounds better—than its predecessor, GPT-2.

Suddenly, the theoretical idea of artificial general intelligence, which performs as well as or better than humans on a wide variety of tasks, seemed tantalizingly close. If the scaling law held, A.I. companies might achieve A.G.I. by pouring more money and computing power into language models. Within a year, Sam Altman, the chief executive at OpenAI, published a blog post titled “Moore’s Law for Everything,” which argued that A.I. will take over “more and more of the work that people now do” and create unimaginable wealth for the owners of capital. “This technological revolution is unstoppable,” he wrote. “The world will change so rapidly and drastically that an equally drastic change in policy will be needed to distribute this wealth and enable more people to pursue the life they want.”

It’s hard to overstate how completely the A.I. community came to believe that it would inevitably scale its way to A.G.I. In 2022, Gary Marcus, an A.I. entrepreneur and an emeritus professor of psychology and neural science at N.Y.U., pushed back on Kaplan’s paper, noting that “the so-called scaling laws aren’t universal laws like gravity but rather mere observations that might not hold forever.” The negative response was fierce and swift. “No other essay I have ever written has been ridiculed by as many people, or as many famous people, from Sam Altman and Greg Brockton to Yann LeCun and Elon Musk,” Marcus later reflected. He recently told me that his remarks essentially “excommunicated” him from the world of machine learning. Soon, ChatGPT would reach a hundred million users faster than any digital service in history; in March, 2023, OpenAI’s next release, GPT-4, vaulted so far up the scaling curve that it inspired a Microsoft research paper titled “Sparks of Artificial General Intelligence.” Over the following year, venture-capital spending on A.I. jumped by eighty per cent…(More)”.

Article by Courtenay Brown, and Emily Peck: “There’s a new fear among investors and CEOs: flying blind on investments without sufficient data on the economy’s health.

Why it matters: The U.S. government produces some of the world’s premiere economic data. The future of those indicators looks murkier than ever, with no private sector source readily available to replace them.

How it works: The Bureau of Labor Statistics, which President Trump wants to overhaul, produces crucial data on jobs and inflation.

• Other sources, including private sector employment data from payroll processor ADP, help shape the understanding of how the economy is performing. But nothing yet can replace traditional government data.

What they’re saying: “At the end of the day all roads lead back to the government data,” Mark Zandi, chief economist at Moody’s Analytics, tells Axios. “If we don’t have that data, we’re going to be lost.”

• Businesses, state and local governments, the Federal Reserve and beyond will substitute private data that might be less reliable and “make decisions that are are just plain wrong when it matters the most,” Zandi said.

The intrigue: E.J. Antoni, President Trump’s pick to lead the Bureau of Labor Statistics, said those fears are the very reason why the agency should release its monthly jobs report less frequently.

• Along with the Trump administration, he’s criticized the BLS for the data revisions in its unemployment report.
• “How on Earth are businesses supposed to plan — or how is the Fed supposed to conduct monetary policy — when they don’t know how many jobs are being added or lost in our economy?” Antoni told Fox prior to his nomination.
• He suggested suspending the jobs report entirely, for an indefinite period of time, while the BLS revised its methods.

The other side: “Blaming the data because you get things wrong is bad. That’s a bad forecaster,” Joe Lavorgna, an economist who counsels Treasury Secretary Scott Bessent, tells Axios.

• “Questioning the data when you get multiple standard deviation missed and biases in the data in one direction — people have the right to question that.”..(More)”.

Article by Jay Peters: “Reddit says that it has caught AI companies scraping its data from the Internet Archive’s Wayback Machine, so it’s going to start blocking the Internet Archive from indexing the vast majority of Reddit. The Wayback Machine will no longer be able to crawl post detail pages, comments, or profiles; instead, it will only be able to index the Reddit.com homepage, which effectively means Internet Archive will only be able to archive insights into which news headlines and posts were most popular on a given day.

”Internet Archive provides a service to the open web, but we’ve been made aware of instances where AI companies violate platform policies, including ours, and scrape data from the Wayback Machine,” spokesperson Tim Rathschmidt tells The Verge.

The Internet Archive’s mission is to keep a digital archive of websites on the internet and “other cultural artifacts,” and the Wayback Machine is a tool you can use to look at pages as they appeared on certain dates, but Reddit believes not all of its content should be archived that way. “Until they’re able to defend their site and comply with platform policies (e.g., respecting user privacy, re: deleting removed content) we’re limiting some of their access to Reddit data to protect redditors,” Rathschmidt says.

The limits will start “ramping up” today, and Reddit says it reached out to the Internet Archive “in advance” to “inform them of the limits before they go into effect,” according to Rathschmidt. He says Reddit has also “raised concerns” about the ability of people to scrape content from the Internet Archive in the past…(More)”.

Paper by Juan Zambrano et al: “Participatory Budgeting (PB) empowers citizens to propose and vote on public investment projects. Yet, despite its democratic potential, PB initiatives often suffer from low participation rates, limiting their visibility and perceived legitimacy. In this work, we aim to strengthen PB elections in two key ways: by supporting project proposers in crafting better proposals, and by helping PB organizers manage large volumes of submissions in a transparent manner. We propose a privacy-preserving approach to predict which PB proposals are likely to be funded, using only their textual descriptions and anonymous historical voting records — without relying on voter demographics or personally identifiable information. We evaluate the performance of GPT 4 Turbo in forecasting proposal outcomes across varying contextual scenarios, observing that the LLM’s prior knowledge needs to be complemented by past voting data to obtain predictions reflecting real-world PB voting behavior. Our findings highlight the potential of AI-driven tools to support PB processes by improving transparency, planning efficiency, and civic engagement…(More)”.

Blog by Mark Headd: “…It’s not surprising that the civic tech world has largely metabolized the rise of Artificial Intelligence (AI) as a set of tools we can use to make these interfaces even better. Chatbots! Accessible PDFs! These are good and righteous efforts that make things easier for government employees and better for the people they serve. But they’re sitting on a fault line that AI is shifting beneath our feet: What if the primacy and focus we give *interfaces, *and the constraints we’ve accepted as immutable, are changing?..

Modern generative AI tools can assemble complex, high-fidelity interfaces quickly and cheaply. If you’re a civic designer used to hand-crafting bespoke interfaces with care, the idea of just-in-time interfaces in production makes your hair stand on end. Us, too. The reality is, this is still an idea that lies in the future. But the future is getting here very quickly.

Shopify, with its 5M DAUs and $292B processed annually, is doing its internal prototyping with generative AI. Delivering production UIs this way is gaining steam both in theory and in proof-of-concept (e.g., adaptive UIs, Fred Hohman’s Project Biscuit, Sean Grove’s ConjureUI demo). The idea is serious enough that Google, not a slouch in the setting-web-standards game, is getting into the mix with Stitch and Opal. AWS is throwing its hat in the ring too. Smaller players like BuildAI, Replit, Figma, and Camunda are exploring LLM-driven UI generation and workflow design. All of these at first may generate wacky interfaces and internet horror stories, and right now they’re mostly focused on dynamic UI generation for a developer, not a user. But these are all different implementations of an idea that are converging on a clear endpoint, and if they can get into use at any substantial scale, they will become more reliable and production ready very quickly…(More)”.

Paper by Eray Erturk et al: “Wearable devices record physiological and behavioral signals that can improve health predictions. While foundation models are increasingly used for such predictions, they have been primarily applied to low-level sensor data, despite behavioral data often being more informative due to their alignment with physiologically relevant timescales and quantities. We develop foundation models of such behavioral signals using over 2.5B hours of wearable data from 162K individuals, systematically optimizing architectures and tokenization strategies for this unique dataset. Evaluated on 57 health-related tasks, our model shows strong performance across diverse real-world applications including individual-level classification and time-varying health state prediction. The model excels in behavior-driven tasks like sleep prediction, and improves further when combined with representations of raw sensor data. These results underscore the importance of tailoring foundation model design to wearables and demonstrate the potential to enable new health applications…(More)”

Paper by Moritz Schütz, Lukas Kriesch, Sebastian Losacker: “The relevance of institutions for regional development has been well established in economic geography. In this context, local and regional governments play a central role, particularly through place-based and place-sensitive strategies. However, systematic and scalable insights into their priorities and strategies remain limited due to data availability. This paper develops a methodological approach for the comprehensive measurement and analysis of local governance activities using web mining, natural language processing (NLP), and machine learning techniques. We construct a novel dataset by web scraping and extracting cleaned text data from German county and municipality websites, which provides detailed information on local government functions, services, and regulations. Our county-level topic modelling approach identifies 205 topics, from which we select 30 prominent topics to demonstrate the variety of topics found on county websites. An in-depth analysis of the three exemplary topics, Urban Development and Planning, Climate Protection Initiatives, and Business Development and Support, reveals how strategic priorities vary across space and how counties differ in their framing of similar topics. This study offers an explanatory framework for analysing the discursive dimensions of local governance and mapping regional differences in policy focus. In doing so, it expands the methodological toolkit of regional research and opens new avenues in understanding local governance through web data. We make an aggregated version of the data set freely available online…(More)”.

Report by National Academies of Sciences, Engineering, and Medicine: “In recent years, Lidar technology has improved. Additionally, the experiences of state departments of transportation (DOTs) with Lidar have grown, and documentation of existing practices, business uses, and needs would now benefit state DOTs’ efforts.

NCHRP Synthesis 642: Practices for Collecting, Managing, and Using Light Detection and Ranging Data, from TRB’s National Cooperative Highway Research Program, documents state DOTs’ practices related to technical, administrative, policy, and other aspects of collecting, managing, and using Lidar data to support current and future practices…(More)”

Article by Madison Leeson: “Cultural heritage researchers often have to sift through a mountain of data related to the cultural items they study, including reports, museum records, news, and databases. The information in these sources contains a significant amount of unstructured and semi-structured data, including ownership histories (‘provenance’), object descriptions, and timelines, which presents an opportunity to leverage automated systems. Recognising the scale and importance of the issue, researchers at the Italian Institute of Technology’s Centre for Cultural Heritage Technology have fine-tuned three natural language processing (NLP) models to distill key information from these unstructured texts. This was performed within the scope of the EU-funded RITHMS project, which has built a digital platform for law enforcement to trace illicit cultural goods using social network analysis (SNA). The research team aimed to fill the critical gap: how do we transform complex textual records into clean, structured, analysable data?

The paper introduces a streamlined pipeline to create custom, domain-specific datasets from textual heritage records, then trained and fine-tuned NLP models (derived from spaCy) to perform named entity recognition (NER) on challenging inputs like provenance, museum registries, and records of stolen and missing art and artefacts. It evaluates zero-shot models such as GLiNER, and employs Meta’s Llama3 (8B) to bootstrap high-quality annotations, minimising the need for manual labelling of the data. The result? Fine-tuned transformer models (especially on provenance data) significantly outperformed out-of-the-box models, highlighting the power of small, curated training sets in a specialised domain…(More)