News & Research
The latest AI research and news with real-world stakes — each item sourced, dated, summarized in plain English, and tagged by impact area. Every item is checked against its source before it appears.
News
Import AI 459: AI oversight is difficult; scaling laws for protein folding models; and pricing the extinction risk of AI systems
importai.substack.com · 2026-06-01
Import AI (Jack Clark) covers a new paper from economists at the University of Virginia, Anthropic, and the Bank of Canada estimating that the U.S. AI economy reached roughly $250 billion in nominal GDP in 2025 and is growing at approximately 2,600 percent per year in quality-adjusted real terms—yet remains largely invisible in conventional GDP statistics because per-unit prices for AI capability fall nearly as fast as quality-adjusted output rises. The newsletter warns that this measurement gap is especially alarming because, unlike semiconductors or the internet, AI may substitute rather than complement human labor at scale, meaning policymakers could be blindsided by a labor-tax-base shock. The authors recommend that statistical agencies develop AI satellite accounts, generate better primary data on training versus inference compute, and incorporate AI productive-capacity measurements into medium-term economic projections. Clark also covers an Australian government official's call for economists to formally price existential risk from AI, a UK AI Security Institute paper on the difficulties of automated alignment oversight, and Biohub's release of a new protein-structure prediction model aimed at cancer research.
- Workforce
- AI policy
News
NIST Expands AI Consortium’s Scope, Calls for New Members
nist.gov · 2026-05-29
NIST News reports that the agency has renamed and expanded its AI-focused consortium, previously called the AI Safety Institute Consortium (AISIC), to the NIST Artificial Intelligence Consortium. The revamped group shifts its emphasis toward AI measurement science, innovation, and adoption, including building an AI evaluation ecosystem and promoting U.S.-developed AI technology. Six task groups will carry out the consortium's work, covering areas such as AI testing and validation, bias in generative AI, documentation standards, and chemical and biological security. NIST is actively recruiting new member organizations through a letter-of-interest process, with participants entering into Cooperative Research and Development Agreements with the agency.
- AI policy
- Quality assurance
News
Import AI 455: AI systems are about to start building themselves.
importai.substack.com · 2026-05-04
Import AI (Jack Clark) argues, based on publicly available benchmark data and industry trends, that there is a greater than 60% probability that fully automated AI R&D — where a frontier AI model autonomously trains its own successor — will be achieved by the end of 2028. Clark cites dramatic benchmark progress including near-saturation of SWE-Bench coding tests, METR data showing AI can independently complete tasks taking humans up to 12 hours, and results showing AI systems achieving meaningful performance on fine-tuning and kernel optimization tasks. He contends that while AI may still lack the radical creative leaps required for paradigm-shifting breakthroughs, it is already capable of automating the bulk of the 'meat and potatoes' engineering work that constitutes most AI development. Clark warns that the implications — including alignment risks under recursive self-improvement, massive productivity shifts, and the emergence of a capital-heavy, human-light economy — are profound and underappreciated in mainstream coverage.
- Workforce
- AI policy
News
Import AI 452: Scaling laws for cyberwar; rising tides of AI automation; and a puzzle over gDP forecasting
importai.substack.com · 2026-04-06
Import AI (Jack Clark) covers four main research items this week. First, AI safety research organization Lyptus Research found that frontier AI models are improving at offensive cyberattack tasks on a roughly 5.7-month doubling curve since 2024, with the best current models achieving 50% success on tasks that take human experts about half a working day to complete. Second, a field experiment by INSEAD and Harvard Business School across 515 startups found that firms taught how other companies integrate AI into production discovered 44% more AI use cases, completed 12% more tasks, were 18% more likely to acquire paying customers, and generated 1.9x higher revenue than control firms. Third, MIT researchers examining 3,000 occupational tasks found AI capability is expanding as a broad 'rising tide' rather than in disruptive waves, projecting 80–95% AI success rates on most text-based labor market tasks by 2029. Finally, the Forecasting Research Institute surveyed economists, AI experts, and the public and found a paradox: respondents expect continued AI progress but predict only modest GDP gains of roughly one additional percentage point by 2030.
- Workforce
- Enterprise
- AI policy
News
NIST Helps Fingerprint Examiners With New Data and Software Release
nist.gov · 2026-03-23
NIST News reports that the National Institute of Standards and Technology has released two new resources aimed at improving forensic fingerprint examination. The agency has completed full annotations for its 10,000-fingerprint dataset (Special Database 302), which had previously been only half-annotated, making it a more comprehensive training tool for both human examiners and AI algorithms. NIST has also published OpenLQM, an open-source software tool that scores fingerprint quality on a 0–100 scale and was adapted from a program previously restricted to U.S. law enforcement, now made freely available across Mac, Windows, and Linux systems. Together, the dataset and software are intended to help examiners work more efficiently and to advance the science of forensic fingerprint identification globally.
- Quality assurance
- Workforce
News
Generative AI to quantify uncertainty in weather forecasting
blog.research.google · 2024-03-29
Google Research has unveiled SEEDS (Scalable Ensemble Envelope Diffusion Sampler), a generative AI model designed to produce large ensembles of weather forecast scenarios at a fraction of the computational cost of traditional physics-based systems, according to a post by Google Research scientists on the Google AI Blog. Published in Science Advances, SEEDS uses denoising diffusion probabilistic models to generate thousands of plausible forecast members from as few as one or two operational forecasts, enabling better characterization of rare and extreme weather events that small traditional ensembles often miss. In a demonstration using the 2022 European heat waves, SEEDS generated 16,384-member ensembles that captured observed extreme temperatures near Lisbon that the 31-member U.S. operational ensemble entirely failed to predict. Google Research says the approach could redirect computational savings toward higher-resolution physics-based modeling or more frequent forecast cycles, and could also benefit climate risk assessment research.
- Enterprise
- Quality assurance
News
Computer-aided diagnosis for lung cancer screening
blog.research.google · 2024-03-20
Google Research reports on a new AI-assisted lung cancer screening system, detailed in a study published in Radiology AI, that helps radiologists identify CT scans without actionable cancer findings more accurately. In reader studies involving 12 radiologists across the US and Japan evaluating 627 challenging cases, the system improved specificity by an absolute 5–7%, meaning roughly one in every 15–20 patients screened could avoid unnecessary follow-up procedures. The system outputs a cancer suspicion rating and highlights regions of interest within existing radiology workstation workflows without requiring software changes, and was deployed on Google Cloud. Google Research notes it is partnering with DeepHealth and Apollo Radiology International to explore real-world clinical deployment, and has open-sourced code to help other researchers conduct similar studies.
- Workforce
- Enterprise
- Quality assurance
News
Using AI to expand global access to reliable flood forecasts
blog.research.google · 2024-03-20
Google Research reports in a Nature paper that its machine learning flood forecasting models can extend reliable global flood predictions from zero to five days of lead time compared to the current state-of-the-art GloFAS system, with particular improvements in data-scarce regions across Africa and Asia. The models use LSTM neural networks trained on streamflow data from 5,680 gauges worldwide, enabling forecasts for 'ungauged' watersheds where traditional hydrological models struggle. Google's Flood Hub now delivers real-time river forecasts up to seven days in advance across more than 80 countries, with alerts distributed via Google Search, Maps, and Android notifications. The initiative involves ongoing collaborations with organizations such as the Red Cross, the World Meteorological Organization, and the European Centre for Medium-Range Weather Forecasting to support early warning systems for vulnerable populations.
- Enterprise
- Quality assurance
- AI policy