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.
In this week’s tech news, Google collaborates with the Wild Dolphin Project (WDP) to build CHAT, a tool that uses DolphinGemma to study dolphin vocalizations. Shweta K, a Threads user who shares AI insights, posted their thoughts on a video demonstrating real-time accent neutralization with Krisp AI Accent Conversion. Anthropic introduces a Research tool and a Google Workspace integration in beta for its AI assistant Claude. New Scientist reports on a study that found that it is becoming increasingly difficult for people to tell the difference between human and AI-generated deepfake voices. OpenAI releases GPT-4.1, a model with similar performance to GPT-4.5 with lower costs and latency, and announces the offboarding of GPT-4.5. OpenAI also releases a prompting guide for GPT-4.1 to help users achieve the best results from the model’s new approach to responding to user prompts. Microsoft’s Azure OpenAI Service offers users access to OpenAI’s models with more security. A new study proposes CaMeL, a defense mechanism to protect LLMs from prompt injection attacks. Finally, Simon Willison explains more about prompt injection on his blog.
Here’s this week’s news:
Google’s AI Model for Dolphin Communication
Google has developed an AI model named DolphinGemma in collaboration with the Wild Dolphin Project (WDP) to work towards understanding and eventually communicating with dolphins. The project aims to decode the structure of dolphin communication by training the model on WDP’s datasets of dolphin sounds. Researchers hope to use CHAT (Cetacean Hearing Augmentation Telemetry) with DolphinGemma to analyze and replicate dolphin vocalizations, eventually expanding to exploring interspecies communication using advanced AI techniques. The model is scheduled for open-access release in the summer of 2025. Read more.
Watch a demonstration of CHAT here.
Commentary on Krisp AI Accent Conversion
Shweta K, a Threads app user who posts their insights on AI, shared a video of a man demonstrating how he uses Krisp AI Accent Conversion to neutralize his Indian English accent. Krisp was designed to reduce background sounds and soften off-shore call center agents’ accents while speaking with clients in real time. Commenters on the post expressed mixed reviews about the tool. View the post.
Similarly, Reuters reported on a 2009 UK study about respondents’ preferred call center accents. Researchers found that over half of respondents preferred the “Queen’s English” when speaking with call center agents and found non-British accents difficult to understand. The study revealed that both gender and accents affected a brand’s consumer appeal. Read more.
Learn more about Krisp AI Accent Conversion here.
Anthropic’s Claude Introduces Research and Workspace Tools
Anthropic launched two major features for its Claude AI assistant: a research tool and Google Workspace integration. The research tool enables Claude to perform multi-step, citation-backed web searches to answer complex queries more accurately. The Google Workspace integration allows Claude to access Gmail, Calendar, and Docs, streamlining tasks like summarizing meeting notes and tracking action items. All paid users of Claude can access the Google Workspace integration in beta, while the Research tool is only available in early beta to Max, Team, and Enterprise plan subscribers. Read more.
Spotting AI Deepfake Voices
A study covered by New Scientist shows that people often fail to distinguish AI-generated voices from real human ones. Even with training, participants misidentified synthetic voices about 40% of the time. This raises concerns over the potential misuse of deepfake audio, particularly in scams and misinformation. The findings suggest urgent need for better detection tools and public awareness. Read more.
GPT-4.1 Release Overview
OpenAI introduced GPT-4.1, a significant upgrade to its previous models with improved performance in reasoning, tool usage, and instruction following. GPT-4.1 is now available in OpenAI’s API with enhanced support for longer contexts up to a 1 million tokens and more accurate responses. It includes better support for coding and multimodal input handling. With the release of GPT-4.1 OpenAI announced that it would be phasing out GPT-4.5 since GPT-4.1 offers a similar performance at a lower cost and latency. Read more.
GPT-4.1 Prompting Guide
The GPT-4.1 Prompting Guide provides best practices for utilizing the model’s new features, particularly in agentic and tool-augmented workflows. It emphasizes the importance of clear and specific instructions, planning prompts, and structured system messages to guide model behavior. Developers are encouraged to use API-native tool formatting for optimal results. The guide highlights improvements in task planning and tool usage, making GPT-4.1 especially useful for complex or autonomous problem-solving tasks. Read more.
Microsoft’s Azure OpenAI Service
The Azure OpenAI Service offers access to OpenAI’s advanced language models, including GPT-4, GPT-3.5, Codex, and DALL·E, through Azure’s secure infrastructure. It supports various tasks such as content generation, summarization, semantic search, and code translation. Users can interact with these models via REST APIs and SDKs in multiple programming languages. The service emphasizes enterprise-grade security, compliance, and responsible AI practices, providing features like virtual network support and content filtering. Read more
Defeating Prompt Injections by Design
A recent study introduces CaMeL, a defense mechanism designed to protect Large Language Model (LLM) agents from prompt injection attacks. CaMeL functions by isolating trusted control and data flows from untrusted inputs, preventing unauthorized influence on the program’s behavior. It also employs capability-based security to restrict data access and prevent leakage. Evaluated using the AgentDojo benchmark, CaMeL achieved provable security in 67% of tasks, demonstrating its effectiveness in enhancing LLM agent security. Read more
Prompt Injection Explained (November 2023 Edition)
Simon Willison provides an accessible explanation of prompt injection, a security vulnerability affecting applications built on Large Language Models. He illustrates how untrusted user inputs can override developer instructions, leading to unintended behaviors. For instance, a user might input a command that causes an AI assistant to perform unauthorized actions. Willison emphasizes that, as of late 2023, there is no definitive solution to this issue, highlighting the need for continued awareness and research in this area. Read more
Tool of The Week: UBC GenAI Toolkit (TypeScript)
The UBC GenAI Toolkit is a modular TypeScript library developed to streamline the integration of Generative AI capabilities into web applications. It provides standardized interfaces for common GenAI tasks, abstracting the complexities of underlying implementations. By following the Facade pattern, the toolkit ensures API stability and simplifies the adoption of new technologies, allowing developers to focus on application logic rather than GenAI infrastructure. Read more
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 […]