College professors across the country are weaving AI tools like Claude and ChatGPT into their teaching routines. At the University of Pittsburgh, instructors use these systems to grade exams, create math problems, write student feedback and draft lecture notes. Some faculty members, however, worry about accuracy and overreliance, and they fact-check AI-generated content before presenting it to students.
The debate around AI in classrooms extends well beyond STEM subjects. Music and English teachers now use AI for lesson planning and rubric creation. At Florida State University, students and professors disagree about how much AI is acceptable. Some students say AI-generated lesson plans feel formulaic and lack depth, pushing faculty to design assignments that resist easy automation, such as oral exams and creative projects.
Meanwhile, concerns about AI evaluation systems have surfaced in a recent study. Researchers auditing public agent leaderboards HAL and tau2-bench found that grading errors silently shifted rankings. Problems included double-counting costs, mislabeling actions and misreading files, which in one case dropped scores by 59 to 82 points. The study calls for simple safeguards like reporting graded fractions and pinning evaluator versions.
On the global stage, physical limits are shaping the AI race. The United States leads in advanced models and data center use, accounting for 45 percent of global data center electricity consumption in 2024. China, however, builds facilities more quickly. New U.S. policies, including the Ratepayer Protection Act and a bulk-power order, aim to manage energy use and reduce reliance on Chinese equipment.
Politics around AI regulation are also shifting. Democrats in Congress plan closer oversight of companies building advanced AI models, drawing a clear contrast to the hands-off approach under President Donald Trump. AI costs and risks became a major issue during the midterm elections, though partisan disagreements and potential vetoes make new legislation unlikely in the near term.
Voters on November 3 further signaled unease with the industry by rejecting data center construction in their communities and electing Democrats focused on AI risk. Colorado Rep.-elect Manny Rutinel will enter Congress with a detailed AI agenda, and Democrats may use oversight powers to investigate Big Tech.
OpenAI has named former Secretary of the Army Dan Driscoll as an adviser, tasking him with guiding senior leadership on using advanced AI for national security and cybersecurity, including work with United States allies. In a separate development, about 150 chief information security officers met at a Microsoft summit to discuss AI agent security. Microsoft's Yina Arenas said the industry is split between those locking down agents and those letting them operate freely, and she identified people, not technology, as the biggest challenge.
Europe's highest court issued a landmark opinion on generative AI and copyright. The CJEU advises that training large language models on copyrighted works may be permitted under EU law, but reproducing protected content in chatbot responses could still lead to infringement. Separately, the 3rd Annual Globee Awards opened nominations worldwide for achievements in generative AI, machine learning, AI agents and other applications, judged through a merit-based process.
Key Takeaways
- University of Pittsburgh professors use Claude and ChatGPT for grading, feedback, lecture notes and math problems, though some worry about inaccuracy.
- Students at Florida State University say AI-generated lesson plans feel formulaic, pushing professors to design AI-resistant assignments.
- Grading errors on leaderboards HAL and tau2-bench silently shifted AI agent rankings, with one error dropping scores by 59 to 82 points.
- The United States accounts for 45 percent of global data center electricity use in 2024, while China leads in rapid facility construction.
- Democrats plan closer AI industry scrutiny, contrasting with President Trump's hands-off approach, though new legislation faces hurdles.
- Voters rejected data center construction on November 3 and elected Democrats focused on AI risk, including Colorado Rep.-elect Manny Rutinel.
- OpenAI hired former Army Secretary Dan Driscoll to advise on AI national security and cybersecurity, including work with U.S. allies.
- Microsoft's AI agent security summit revealed a split between locking down agents and allowing free operation, with people identified as the main challenge.
- The CJEU says training large language models on copyrighted works may be allowed under EU law, but reproducing protected content in responses could still infringe.
- The 3rd Annual Globee Awards accept worldwide nominations for generative AI, machine learning and AI agent achievements.
Pitt professors use AI tools like Claude for teaching and grading
University of Pittsburgh professors are using AI tools such as Claude and ChatGPT to help with tasks like creating math problems, writing feedback and taking lecture notes. Anne Yust uses AI to grade exams and give student feedback. Maggie Jones uses AI for brainstorming but avoids using it for writing. Weifeng Xu uses AI to create lecture notes but fact-checks them first. Some professors worry about AI inaccuracy and overreliance.
College professors debate responsible AI use in classrooms
Artificial intelligence is spreading into college classrooms beyond STEM subjects, with music and English teachers using AI for lesson planning and rubric creation. Florida State University students and professors disagree about how much AI is acceptable. Some students say AI-generated lesson plans feel formulaic and lack depth. Professors are urged to be honest about their AI use and to design assignments that cannot be easily completed by AI, such as oral exams or creative projects.
Grading errors silently alter AI agent leaderboard rankings
A study found that errors in grading software can silently change the rankings of AI agent leaderboards without anyone noticing. Researchers auditing public leaderboards HAL and tau2-bench found problems like double-counting costs and mislabeling actions. One fix that was announced actually reordered a leaderboard. A mode check misread four files and dropped scores by 59 to 82 points without reporting an error. The study recommends simple checks like reporting graded fractions and pinning evaluator versions.
Power and chip limits shape the global AI race
The race to lead in artificial intelligence is limited by physical barriers like electricity shortages, chip supply and local opposition to data centers. The United States leads in advanced models and data center use, accounting for 45 percent of global data center electricity use in 2024. China leads in building facilities quickly but faces different challenges. New U.S. policies including the Ratepayer Protection Act and a bulk-power order aim to manage energy use and reduce reliance on Chinese equipment.
Democrats plan closer AI industry scrutiny but rules face hurdles
Democrats in Congress plan to increase oversight of companies building advanced artificial intelligence models as a contrast to President Donald Trump's hands-off approach. AI costs and risks became an important issue during the midterm elections. However, Democrats face challenges in agreeing on how to regulate AI. Potential vetoes from Trump make it unlikely that new AI legislation will pass soon.
OpenAI hires former Army secretary to advise on AI and cybersecurity
OpenAI named former Secretary of the Army Dan Driscoll as an adviser. He will guide senior leadership on using advanced AI for national security and cybersecurity. His focus will include working with United States allies.
3rd Annual Globee Awards open AI achievement nominations worldwide
The Globee Awards announced nominations for the 3rd Annual Globee Awards for Artificial Intelligence. Organizations worldwide can submit achievements involving generative AI, machine learning, AI agents, and other AI applications. Entries are judged through a merit-based process with scoring criteria set by industry professionals.
Midterms shift AI politics as voters reject data centers and back AI regulation
Voters on November 3 rejected data center construction in their communities and elected Democrats focused on AI risk. AI industry super PACs struggled while AI-safety groups backed candidates who won key races. Colorado Rep.-elect Manny Rutinel will enter Congress with a detailed AI agenda, and Democrats may use oversight powers to investigate Big Tech.
EU top court opinion may allow LLM training but warns about copyright in responses
Europe's highest court issued a landmark opinion on generative AI and copyright. The CJEU advises that training large language models on copyrighted works may be allowed under EU law. However, reproducing protected content in chatbot responses could still lead to copyright infringement.
Microsoft summit explores how AI agent security mirrors managing people
About 150 chief information security officers met at a Microsoft summit to discuss AI agent freedom. Microsoft's Yina Arenas described a split in the industry, with some leaders locking down agents and others letting them operate freely. She said the biggest challenge is not the technology but the people who use it and are responsible for AI agent actions.
Sources
- āAlmost like a colleague sitting next to meā: Pitt professors describe how they use AI
- AI in classrooms ā how can you āresponsiblyā use it?
- Silent Instrument Errors Reorder Agent Leaderboards
- Building around the limits of the AI raceĀ
- AI Giants to Come Under Fresh Scrutiny But New Rules Unlikely
- OpenAI hires former Army secretary as adviser on national security, cybersecurity
- The 3rd Annual GlobeeĀ® Awards for Artificial Intelligence Invite Product, Service, and Business Achievement Nominations Worldwide
- AI & Tech Brief: Midterms turn the AI tide
- Landmark CJEU opinion considers allowing LLM training, with AI opt-out
- Why AI Agent Security Looks a Lot Like Managing Employees
Comments
Please log in to post a comment.