What You Should Know
If you only do one thing with this newsletter, watch this video. Google’s Veo 3 AI video generator is the most realistic AI video generation available, with easily attainable videos and images nearly-indistinguishable from reality. With incredibly sophisticated scams becoming more common and accessible to bad actors, one of the most important skills that can be built today is a highly discerning approach to media authenticity. (Full paper by OpenAI referenced in NY Post Article).
In addition to bad actors becoming bolder and more convincing, privacy concerns are rising now that OpenAI is now required by court order to retain all user content indefinitely. Deleted chats, temporary chats, and accounts that have opted-out of their information training GPT models will not prevent OpenAI from retaining this data.
AI In Education
- Clay Shirky, vice provost for AI and technology in education at NYU, wrote an insightful article and gave a follow-up podcast about the fruits, struggles, and concerns of AI integration at NYU among both faculty and students. Highly recommended.
- The Ohio State University is launching an AI Fluency initiative, incorporating AI into General Education. AI will be integrated into existing coursework and seminars, with additional workshops and GenAI courses being available for all majors. Additional support is given to faculty and staff as well, a critical component to productively build AI use into the classroom.
- Enrollment fraud has skyrocketed as colleges, in part due to scammers frequently using AI chatbots and targeting online courses. In 2024, California alone reported 1.2 million fraudulent applications, resulting in 223,000 suspected fake enrollments and $11.1 million in financial aid stolen.
AI In Research
- The first AI system has independently passed peer review at an A* scientific conference, ACL 2025. Zochi, an “AI Scientist”, autonomously conducted the full scientific method through peer-review and publication. According to the researchers, humans only supplied the general research domain while Zochi “identified the research direction, formalized the method, implemented, tested, conducted experiments, and wrote/presented the paper (excluding figures/formatting)”.
- Apple researchers published a paper stating that large reasoning models face a “complete accuracy collapse” when faced with highly complex problems. The paper has been criticized by some as Apple attempting to mitigate fallout of their fumbling the delivery of “Apple Intelligence” and by remaining far behind in the AI race.
- A recent study by MIT and OpenAI found that heavy users of AI as an emotional companion had the most negative outcomes, including loneliness, less socialization with real people, and more emotional dependence on the chatbot. The top 10% of users (by usage time) were 3x as likely to feel distress if ChatGPT was unavailable.
- Six widely used LLMs outperformed the average human scores on five different Emotional Intelligence tests with an average accuracy of 81% compared to the 56% human average. ChatGPT-4 was also able to generate new items that demonstrated statistically equivalent test difficulty, “[suggesting] that LLMS can generate responses that are consistent with accurate human knowledge about human emotions and their regulation.”
AI Current Events
- There are currently over 1,000 AI Bills currently under consideration by state legislature. This comes as the U.S. Senate considers the One Big, Beautiful Bill that includes a provision prohibiting state regulation of AI for 10 years. State lawmakers are pushing back with more than 260 state legislators from all 50 states opposing the provision.
- Anthropic CEO Dario Amodei predicts that 50% of all entry-level white collar roles could disappear within 5 years due to AI. Ravi Kumar, CEO of Cognizant (a 350,000-person IT company), on the other hand argues that AI create more jobs for new graduates, not less.
- The Trump administration plans to reorganize the U.S. AI Safety Institute (AISI) into a new body called the Center for AI Standards and Innovation (CAISI). The organization will still be a part of the National Institute of Standards and Technology (NIST), and will “lead unclassified evaluations of AI capabilities that may pose risks to national security”.
Weekly AI Tip
Interested in creating a Socratic tutor with an AI Chatbot? The Socratic method employs probing and leading questions to help guide the learner to the answer, without delivering the answer directly. Vinay Ramasesh, research scientist from Google’s DeepMind, provided this prompt to instruct an AI Chatbot to operate in this manner:
“I would benefit most from an explanation style in which you frequently pause to confirm, via asking me test questions, that I’ve understood your explanations so far. Particularly helpful are test questions related to simple, explicit examples. When you pause and ask me a test question, do not continue the explanation until I have answered the questions to your satisfaction. I.e. do not keep generating the explanation, actually wait for me to respond first. Thanks!”
AI Spotlight
Adam Kiehl, a health data scientist at the Veterinary Teaching Hospital (VTH), in collaboration with researchers at the VTH and the CSU Data Science Research Institute, has recently engaged in a research project aimed at developing an AI tool that can be used to accelerate veterinary diagnosis identification from free-text clinical notes. Clinical coding is the process of representing electronic health records with codes from standardized medical terminologies. Coded records can facilitate enhanced research, data sharing, and practice management; however the manual coding process is both time intensive and laborious. The developed AI tool can process free-text clinical notes and automatically assign diagnosis codes, improving the quality of the medical records and accelerating the coding process. The tool was developed by fine-tuning pretrained LLMs on a coded dataset curated by the VTH medical records team. Research into model performance and responsible deployment of the tool is ongoing. For more information, please see the research team’s first paper (https://arxiv.org/abs/2410.15186).
Are you using AI in innovative ways and want to share with your colleagues through this newsletter? Email Brian Kelly at [email protected]
Upcoming AI Events
Essentials of Generative AI Workshop – June 13, 10:00AM-2:00PM in DMC 101
- The Essentials of Generative AI Workshop offers a foundational understanding of Gen AI, covering common terminology, its capabilities, and limitations. The session outlines policies and resources available to faculty and staff. Additionally, the workshop focuses on developing prompting skills and strategies for problem-solving with AI. RSVP
Building Custom GPTs & Gems Workshop – August 8, 10:00AM-2:00PM in DMC 101
- The Building Custom GPTs and Google Gems Workshop focuses on creating personalized and functional AI tools using ChatGPT and Google Gemini. Participants will learn to develop AI tutors, AI generators for assessment content, and AI writing/editing assistants. The workshop is designed so each attendee leaves with their own functional custom AI and the foundational knowledge to build additional tailored solutions. RSVP
AI Disclosure: The following AI tools were utilized in the design and content of this newsletter: ChatGPT o3, Google Gemini 2.5 Pro, Claude Sonnett 4, NotebookLM+
