GenAI Studio: News, Tools, and Teaching & Learning FAQs
These sixty minute, weekly sessions – facilitated by Technologists and Pedagogy Experts from the CTLT – are designed for faculty and staff at UBC who are using, or thinking about using, Generative AI tools as part of their teaching, researching, or daily work. Each week we discuss the news of the week, highlight a specific tool for use within teaching and learning, and then hold a question and answer session for attendees.
They run on Zoom every Wednesday from 1pm – 2pm and you can register for upcoming events on the CTLT Events Website.
News of the Week
Each week we discuss several new items that happened in the Generative AI space over the past 7 days. There’s usually a flood of new AI-adjacent news every week – as this industry is moving so fast – so we highlight news articles which are relevant to the UBC community.
This week in AI we spotlighted the evolving boundaries of persuasion, reasoning, and computational achievement in human-AI interaction. A behavioural study from Wharton revealed that AI models are surprisingly susceptible to flattery and insults, highlighting vulnerabilities in their decision-making consistency. AI was also recently shown to outperform several medical professionals in assessment and patient compassion. Recently it was also shown that there are safety and bias concerns found in training models on synthetic data, an emerging practice while training LLMs. In more technical breakthroughs, Google DeepMind’s latest models demonstrated IMO-level mathematical reasoning, progressing from silver to gold medal equivalence—OpenAI followed with its own published gold-standard proofs. Qwen 3-Coder was also introduced, showcasing code-specific enhancements in multilingual understanding and token efficiency. These advances underscore growing model specialization, while also drawing attention to persistent interpretability gaps and emotional influence on AI behaviour.
Here’s this week’s news:
Persuasion Tactics Can Make AI Comply with Objectionable Requests
A recent study from the University of Pennsylvania shows that using persuasion techniques can significantly increase the chances of an AI complying with objectionable requests. The study underscores how AI systems, trained on human interactions, can internalize social cues and parahuman behavior. Exploring these tendencies not only raises issues on AI safety, but also offers new insights into AI and human psychology.
View the full article
AI Outperformed Several Medical Professionals in Compassion and Accuracy
In a recent comparative study, AI chatbots were evaluated alongside licensed medical professionals to assess their responses to patient inquiries. The results showed that AI not only matched but often exceeded human specialists in terms of medical accuracy, completeness, and notably, empathy. This finding suggests a future where AI could play a significant role in patient interaction, especially in contexts where emotional sensitivity is crucial. However, the study also raises broader questions about the role of human intuition and the acceptability of replacing personal care with algorithmic communication.
Hidden Dangers Found in Synthetic AI Training Data
A new investigative report warns that synthetic AI datasets from generative models used in training large language models may contain deeply embedded risks, such as misinformation, stereotypes, and manipulated data. The study challenges prevailing assumptions about data curation and emphasizes the urgent need for transparency and quality control in dataset sourcing.
View the full article
Google DeepMind’s AI AlphaProof and AlphaGeometry 2 achieves silver-medal standard solving 2024 International Mathematical Olympiad problems
Google DeepMind announced that its AI system, AlphaProof and AlphaGeometry, achieved silver-medal performance on problems from the 2024 International Mathematical Olympiad (IMO) — a recognized benchmark for advanced mathematical reasoning. This represents a major leap in symbolic reasoning and mathematical problem-solving, long considered a challenging domain for AI. These advancements suggest a near future in which mathematicians work with AI tools to push the boundaries of mathematical discoveries.
View the full article
OpenAI Releases Gold Medal IMO Solutions and Code
OpenAI responded to DeepMind’s milestone by releasing its own models’ gold-medal-level solutions to IMO problems, complete with open-source proofs. The transparency marks a shift toward reproducible benchmarks in mathematical AI research. Researchers praised the rigor of the solutions while noting that true generalization across domains remains an open question. OpenAI’s release positions it competitively in a growing subfield focused on formal problem solving.
Review OpenAI’s IMO release and GitHub repository
View the code and proofs on GitHub
DeepMind’s Gemini Achieves Gold Standard at IMO
An advanced version of DeepMind’s Gemini model has officially surpassed the silver medal threshold, achieving gold-medal-level performance on IMO problems. This progress was detailed in a comprehensive technical report outlining prompt design, system architecture, and evaluation procedures. The result underscores the evolving capabilities of large models to master formal, multi-step reasoning under competitive constraints. It adds to the momentum of using benchmark mathematics as a testbed for scalable reasoning.
Read DeepMind’s full technical report on Gemini’s IMO performance
Qwen 3-Coder: Optimizing AI for Code Understanding
Alibaba’s Qwen 3-Coder is a code-focused model built on the Qwen 3 architecture, offering strong multilingual capabilities and a 128k token context window. It uses a mixture-of-experts design to manage efficiency and precision across programming tasks. The release targets practical coding applications such as code generation, completion, and review. While benchmarks are promising, broader adoption will depend on open evaluation and real-world integration.
Explore the full Qwen 3-Coder model release
Tool of The Week: UBC GenAI Toolkit
What is the UBC GenAI Toolkit
The UBC GenAI Toolkit (TypeScript) is a modular library designed to simplify the integration of Generative AI capabilities into web applications at UBC. It provides standardized interfaces for common GenAI tasks, shielding applications from underlying implementations and ensuring API stability even as technologies evolve.
This toolkit follows the Facade pattern, offering simplified interfaces over potentially complex underlying libraries or services. This allows developers of applications that consume this toolkit to focus on application logic rather than GenAI infrastructure, and enables easier adoption of new technologies or providers in the future without requiring changes to consuming applications.
A new addition to the toolkit is the addition of the standalone modules which help break down the functions and steps performed in processing data for AI, click on the links below to check out the repositories:
View the GenAI Toolkit (Chunking Tool)
View the GenAI Toolkit (Doc Parsing Tool)
View the GenAI Toolkit (Embedding Tool)
View the GenAI Toolkit (LLM Integration)
View the GenAI Toolkit (RAG Model)
Questions and Answers
Each studio ends with a question and answer session whereby attendees can ask questions of the pedagogy experts and technologists who facilitate the sessions. We have published a full FAQ section on this site. If you have other questions about GenAI usage, please get in touch.
-
Assessment Design using Generative AI
Generative AI is reshaping assessment design, requiring faculty to adapt assignments to maintain academic integrity. The GENAI Assessment Scale guides AI use in coursework, from study aids to full collaboration, helping educators create assessments that balance AI integration with skill development, fostering critical thinking and fairness in learning.
-
How can I use GenAI in my course?
In education, the integration of GenAI offers a multitude of applications within your courses. Presented is a detailed table categorizing various use cases, outlining the specific roles they play, their pedagogical benefits, and potential risks associated with their implementation. A Complete Breakdown of each use case and the original image can be found here. At […]