Center for Security and Emerging Technology (CSET)

In CSET's June policy.ai newsletter, Owen J. Daniels covered several major stories from the past month. 1️⃣ Both Meta & OpenAI reported efforts to crack down on propaganda networks using AI-generated profiles and content to shape domestic and foreign political opinion, an important topic in a year of many global elections 2️⃣ AnthropicAI & OpenAI claimed to have made progress on work in mechanistic interpretability, trying to pry open the black box of neural networks to support better human understanding of their inner workings 3️⃣ California’s SB 1047 proposed AI safety legislation for frontier models has engendered strong reactions from both tech developers and safety advocates 4️⃣ Palantir has been contracted to build the Pentagon’s data-sharing ecosystem in support of Combined Joint All-Domain Command and Control, one of the Defense Department's highest-profile AI initiatives. Check out these stories in greater detail, as well as the latest research and analysis from CSET, at: https://lnkd.in/gqap7Utf

🚀🚀🚀 Introducing PARAT, our new hub for data about private-sector companies and their AI activities! Explore corporate AI research, patenting, and hiring trends now at https://parat.eto.tech 🚀🚀🚀 PARAT brings together data on hundreds of leading companies around the world and their AI activities, from Big Tech titans and leading generative AI startups to the entire S&P 500. Use the flexible interface to customize your analysis - or crunch the numbers yourself with the export and bulk download features. PARAT's built on our original Private-Sector AI Indicators dataset, also released today - read all about it here: https://lnkd.in/grkfkFQz As always, we're glad to help - visit our support hub (https://eto.tech/support) to contact us, book live support with an ETO staff member or access the latest documentation. 🤖

🗓️ Deadline Approaching! CSET seeks an organized and enthusiastic Strategic Engagement Specialist. Ideal candidates will have a strong development background, and excellent communication skills. Join us and play a key role in ensuring CSET meets its annual fundraising goals!!  Apply by Monday, June 24th. https://lnkd.in/ek9a-q3V

⭐ New CSET Data Brief ⭐ While the U.S. Department of Defense’s broad investments in research and development go far beyond what is publicly disclosed, authors affiliated with the DOD do publish papers about their research. Our new analysis by Emelia Probasco and Autumn Toney examines more than 100,000 papers by DOD-affiliated authors since 2000 and offers insight into the patterns of research publication and collaboration by the DOD. Explore the full report (or read the key takeaways) here: https://lnkd.in/eU9cYm2U

✨New CSET Report✨ China’s leadership believes that artificial intelligence will play a central role in future wars. To understand this complex landscape, CSET’s Sam Bresnick conducted a comprehensive review of dozens of Chinese-language journal articles about AI and the future of warfare, which revealed that Chinese defense experts assert that Beijing is facing several technological challenges that may hinder its ability to capitalize on the advantages provided by military AI. Our new report outlines these perceived barriers and identifies several technologies that Chinese experts believe may help the country develop and deploy military AI-enabled systems. Explore the full report (or read the key takeaways one-pager) here: https://lnkd.in/eSXYEJXQ

✨New CSET Data Snapshot✨ Huawei’s Ascend 910B chip—its latest domestically designed and manufactured AI chip—is challenging one of NVIDIA’s most advanced chips. CSET researchers, Jacob Feldgoise and Hanna Dohmen, dive deep into technical documentation to provide new insights on Huawei’s AI chip development. Their work shows that while the Ascend 910B is an upgrade in performance from earlier versions of the chip, the improvement is not as substantial as advertised. Explore it in full here: https://lnkd.in/eXMVSCAw

⭐ New CSET Report ⭐ Machine learning systems are often unreliable. One intuitively appealing way to handle this problem would be to train systems that “know what they don’t know”—that is, systems that can assess when their own outputs may not be correct. This is known as “uncertainty quantification” and is a challenging research problem within the field of machine learning. Our new explainer by Tim G. J. Rudner and Helen Toner—the fifth installment in our “Key Concepts in AI Safety” series—explores how uncertainty quantification works, why it is so difficult, and what the prospects are for the future. Explore it in full (or read our Key Takeaways one-pager) here: https://lnkd.in/en6R4yiP

📢 Exciting news! We've extended the deadline to apply for our Strategic Engagement Specialist role. Join us and help shape CSET's development strategy and engagement with key external stakeholders. Apply by June 24th, 2024, to join our growing team.  https://lnkd.in/ek9a-q3V #strategicengagement #fundraising #developmentjobs

⭐ New CSET Data Snapshot ⭐ In February 2024, CSET introduced its cybersecurity jobs dataset, a novel resource comprising ~1.4 million LinkedIn profiles of current U.S. cybersecurity workers. Our new data snapshot, published today, by Maggie Wu and Brian Love uses that dataset to identify top-producing institutions of cybersecurity talent. Check it out here: https://lnkd.in/e3abX8aR

One week away! Don't miss it — register now ⬇️