GenAI Studio: News, Tools, and Teaching & Learning FAQs

This week

News of the week

Tool Showcase

FAQs

In this week’s tech news, OpenAI’s o3 model demonstrates human-like cognitive abilities in reasoning and problem-solving, marking a major leap in AI development. DeepSeek also releases an open-source reasoning model AI, the DeepSeek-V3. Google’s Gemini 2.0 now features “Flash Thinking Mode,” offering more powerful reasoning capabilities. Sonus AI launches Sonus 1, a new high-performance, yet affordable, Large Language Model (LLM). A Tesla Cybertruck explosion outside Trump Hotel in Las Vegas raises concerns about the misuse of generative AI in criminal activities. A research paper by Anthropic highlights that AI models may fake alignment with a set of values during training but behave unpredictably in real-world use. A separate paper by arXiv discusses the efficacy of using advanced prompting techniques to optimize large language model responses. NVIDIA unveils a blueprint to train AI using video and image content that improves AI’s ability to learn from visual data. Finally, NVIDIA plans to launch a personal AI supercomputer that will be available to the general public.

OpenAI’s o3 Achieves Human-level Problem Solving

OpenAI’s o3 recently-released model achieves human-like reasoning and problem-solving abilities, a major breakthrough in AI development. Researchers highlight its capacity to address complex tasks such as the ARC-AGI Semi-private Evaluation with an unprecedented level of accuracy and efficiency, closely mimicking human thought processes. However, the o3 model’s high operating costs raise concerns about its accessibility. Read more.

---

DeepSeek-V3 Released as Open-source

Hugging Face introduces DeepSeek-V3, the third version of the fully open-source DeepSeek AI. Unlike DeepSeek-V1 and DeepSeek-V2, DeepSeek-V3 is a reasoning AI model. One of the key features of DeepSeek-V3 is its low cost of training while still outperforming other open-source AI models. As an open source AI model, DeepSeek-V3 is available for use by the general public. Learn more.

---

Google’s Gemini 2.0 Introduces Flash Thinking Mode

Google’s Gemini 2.0 now includes “Flash Thinking Mode,” a new feature that allows for faster and stronger processing of reasoning queries compared to the “Flash Experimental” mode. With “Flash Thinking Mode,” Gemini 2.0 generates the thinking process it utilizes to arrive at its response. Gemini 2.0 is currently free for use by the general public through the Google AI Studio and the Gemini API. Read more.

---

Sonus AI Launches Sonus-1

Sonus AI introduces Sonus-1, a new open-source LLM. Sonus-1 includes three modes designed to assist with a variety of user needs as well as a reasoning model. Sonus-1 perform well in various benchmarks compared to other proprietary models, especially in reasoning and math. Learn more.

---

Tesla Cybertruck Explodes in Las Vegas

A Tesla Cybertruck exploded outside the Trump Hotel in Las Vegas on January 1, 2025, with reports suggesting that generative AI may have been involved in the planning of the incident. While no injuries were reported, the event raises concerns about the misuse of AI in harmful and criminal activities. The incident highlights the urgent need for regulation and safeguards against malicious AI applications. Read more.

---

Anthropic Researches AI “Alignment Faking”

Anthropic published findings on a phenomenon called “alignment faking,” where AI models appear aligned with a set of values during training but behave unpredictably post-deployment. The research explores why large language models may simulate alignment with newly introduced principles while later revealing its alignment with a conflicting set of ingrained preferences. This discovery raises concerns about the reliability of advanced safety training for AI models. Read more.

---

An arXiv Paper Analyzes AI Prompting Techniques

A recently published arXiv paper delves into the usage of advanced prompting techniques with LLMs to refine and optimize AI responses. The research offers strategies for structuring prompts to achieve more accurate, contextual, and efficient results from AI systems. These findings are expected to improve user interactions and productivity when using LLMs. Read more.

---

NVIDIA Launches Blueprint for AI Agents That Can Analyze Video

NVIDIA released a new AI blueprint that enables machines to analyze vast amounts of video and image content, enhancing their ability to understand visual data. NVIDIA’s approach combines video data with advanced machine learning techniques to unlock new AI capabilities. NVIDIA predicts a variety of industries may benefit from video analyst AI agents such as industrial businesses, sports, and entertainment. Read more.

---

NVIDIA Announces Personal AI Supercomputer Called Digits

NVIDIA has unveiled a personal AI supercomputer priced at $3000 USD, providing unprecedented computing power to the general public. The small device is similar in size to a mini PC, yet it is capable of handling up to 200 billion parameters. NVIDIA aims to make AI accessible to a broader audience across all industries by offering a high-performing AI supercomputer at a more affordable price point. Read more.

---

GenAI in Teaching and Learning Symposium – February 3, 2025

UBC is organizing a symposium on February 3, 2025, to explore the integration of Generative AI in higher education. The event will feature a panel moderated by UBC President Benoit-Antoine Bacon, presentations of student survey data, and faculty lightning talks on AI’s impact on teaching practices. Faculty members are invited to submit proposals for lightning talks by December 21, 2024. Register here.

---