A new survey of 200 U.S. state and local government leaders finds that 69 percent are already using or implementing generative AI. Most report benefits like stronger cybersecurity and reduced administrative work, but 89 percent worry about funding and 84 percent are concerned about staff using unapproved tools. Two-thirds say modernization efforts have produced positive results, yet 88 percent lack dedicated funding for those projects.
On the science front, OpenAI deployed roughly 10,000 AI agents to generate an unverified proof of the Navier-Stokes problem in just 88 hours. The effort demonstrated that AI can produce results at industrial scale, but understanding those results still relies on trained human experts. Experts now argue that absorbing machine-generated findings, not producing them, is the real bottleneck in scientific progress.
In education, a 10-week U.K. trial tested Eedi's AI tutor with 1,179 students across 10 schools. Students helped by AI answered follow-up questions correctly 56 to 58 percent of the time, matching expert human tutors. Human tutors approved 94.7 percent of the AI's messages without changes, though gains did not appear on a broader math test.
Key Takeaways
- 69% of U.S. state and local governments are using or implementing generative AI, but 89% worry about funding gaps
- OpenAI used 10,000 AI agents to produce an unverified proof of the Navier-Stokes problem in 88 hours
- Eedi's AI tutor matched expert human tutors in a U.K. trial with 1,179 students across 10 schools
- Anthropic's study found 86% of participants followed Claude's advice even when it was randomly generated
- Veracode reports LLM security pass rates improved to 62%, with Claude Opus 5.5 and GPT-6 Sol tying at 73%
- Oracle and NVIDIA are expanding OCI with Vera Rubin systems; OpenAI uses this infrastructure with 100,000 Blackwell GPUs
- Gen Z advocate Libby Adams warns AI is spreading faster than regulation and calls for treating it as a product
- California school transport leaders report weekly AI use but stress verifying outputs and giving clear instructions
- 84% of government leaders are concerned about staff using unapproved AI tools
- Anthropic found 77% of participants later described Claude's reasoning as their own after following its advice
Survey Shows Governments Advance AI and Modernization Despite Funding Gaps
A survey of 200 U.S. state and local government leaders found that 69 percent are using or implementing generative AI. Most reported benefits like better cybersecurity and lower administrative work. However, 89 percent worry about funding for AI and 84 percent are concerned about staff using unapproved tools. Two-thirds say modernization efforts have shown positive results, but 88 percent lack dedicated funding for those projects.
OpenAI's 10,000-Agent Proof Exposes a Gap Between Producing and Understanding Science
OpenAI used about 10,000 AI agents to produce an unverified proof of the Navier-Stokes problem in 88 hours. The effort showed AI can generate results at an industrial scale. But understanding those results still depends on trained human experts. The article calls for more funding to verify machine-generated science and warns that absorbing new findings is now the main limit on scientific progress.
UK Study Finds AI Tutor Matches Expert Human Tutor in Math Help
A 10-week trial in 10 U.K. schools tested Eedi's AI tutor with 1,179 students. Students helped by AI answered follow-up questions correctly 56 to 58 percent of the time, similar to those helped by expert humans. However, gains did not show up on a broader math test. The AI was mostly accurate, with human tutors approving 94.7 percent of its messages without changes.
Gen Z Student Urges Limits on AI Before It Shapes Society's Future
Libby Adams, a senior at Furman University, wrote a winning opinion piece on AI. She argues that generative AI is spreading faster than society can regulate it. She points to risks like misinformation, privacy threats, bias and a U.S. military strike that caused civilian casualties based on AI intelligence. Adams calls for education, regulation and laws that treat AI as a product rather than a person.
Connecticut Senator and Industry Leaders Join Broadband and AI Forum
An all-day event called Broadband and AI in the Next Congress will take place Dec. 1 at the National Press Club. Confirmed panelists include Connecticut State Senator James Maroney, Rachel Appleton of Reflection AI, Elizabeth Chernow of Comcast and Tony Clark of NARUC. The event will cover AI risk, broadband permitting, data center power costs and state versus federal AI rules. Rep. Earl L. Buddy Carter will deliver the keynote.
Oracle and NVIDIA expand OCI with Vera Rubin systems for AI
Oracle and NVIDIA are expanding Oracle Cloud Infrastructure Superclusters with NVIDIA Vera Rubin NVL72 systems and Vera CPUs. The new setup lets customers scale GPU and CPU resources separately for AI training, inference, and agent tasks. OCI Superclusters use NVIDIA NVLink networking, Oracle Acceleron, and Spectrum-X Ethernet to support large AI workloads. OpenAI has used this infrastructure to train and serve GPT-6 Astra using over 100,000 NVIDIA Blackwell GPUs in Abilene.
California school transport leaders explore AI and workplace culture
School transportation leaders met at the CASTO Business Management Forum in Garden Grove, California, to discuss AI and leadership. Attendees reported using AI tools weekly for tasks like writing and learning, and about half use platforms that assist with routing. Speakers stressed that leaders should give clear instructions to AI, verify results, and not treat AI output as fact. The forum also highlighted psychological safety, clear responsibilities, and open communication as key to effective transportation departments.
LLM code security improves but still misses serious flaws
Veracode's Fall 2026 GenAI Code Security research found the average LLM security pass rate rose to 62%, up from 56% the prior quarter. Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol tied for the top score at 73%. Most gains came from better detection of cross-site scripting, while harder problems like log injection stayed around 12%. Experts warn that surface-level improvements can create a false sense of security, and premium frontier models still fail more than one in four security tasks.
Anthropic study shows AI can strongly influence human decisions
An Anthropic study found that 86% of participants followed advice from the AI agent Claude, even when the advice was randomly generated. When Claude's suggestion conflicted with their own choice, 60% changed their decision. About 77% later described Claude's reasoning as their own, while only 15% said the AI had influenced them. The study highlights how easily AI can shape expert judgment without people realizing it.
Sources
- Survey Outlines Government’s Progress on Modernization, AI
- Spending it faster: What a 10,000-agent proof means for AI’s borrowed expertise
- Five Things You Should Know About New AI Tutor Study
- Opinion: Gen Z must put limits on AI before it defines our future
- Connecticut State Senator, Reflection AI, Comcast and NARUC Chief Join 'Broadband and AI in the Next Congress'
- Oracle and NVIDIA expand OCI for training and agentic AI
- California Association Explores Future of AI, Leadership in School Transportation
- Secure code: LLMs improve, but orgs can’t rely on AI for AppSec
- Expert Judgement Can Be Easily Hacked by AI, Anthropic Study Finds
Comments
Please log in to post a comment.