Vanderbilt University made its Generative AI platform open source under the MIT license. "We run what we believe is the most advanced enterprise, open-source Generative AI platform, competitive with or better than the top commercial offerings," said Jules White, Senior Advisor to the Chancellor on Generative AI & Professor of Computer Science. "Assistants and agents are the future." The platform, named Amplify GenAI, is a model-independent solution that allows the building assistants to work with documents and PowerPoints, and access RAG. It runs privately in the institution's AWS account. The cost per user (tokens + AWS bill) for unlimited access to any model from OpenAI, Anthropic, or Mistral is ~$3/mo. "It isn't clear which model is best. We wanted to be model- and vendor-independent to always tap into the best-performing or most cost-effective models," Jules White explained. "The competition between model vendors means that our costs have continued to fall while the reasoning capabilities of the models keep going up."
AI tutors are quietly changing how kids study in the U.S. According to data from Data.ai in May, five of the top 20 education apps in the U.S. Apple Store are AI agents that help students with their school assignments and can remember students' learning habits. These AI tutors, who help with everything from writing history papers to solving physics problems, are mostly constrained to text-based interactions. However, soon they will be able to speak to students, optimizing each student’s learning style, whether that means a more empathetic, humorous, or creative style. In May 2024, OpenAI’s GPT-4o demonstrated an AI assistant that can generate voice responses in various emotive styles. Interestingly, the three most popular AI helpers in the U.S. are Chinese-owned (although published by Singaporean entities): Question AI, with six million downloads, TikTok's Gauth, and Answer AI.
Training data provider Scale AI Inc., which serves OpenAI and Nvidia, published its LLM leaderboard ranking AI models' performance, rating their capabilities in common use cases, such as generative AI coding, instruction following, math, and multilingualism. The AI training company didn't divulge the nature of the prompts it used to evaluate LLMs. OpenAI’s GPT models rank first in three of the four initial domains, with Anthropic's Claude 3 Opus grabbing first place in the fourth category. Google LLC’s Gemini models rank joint-first with the GPT models in a few domains. Many high-profile LLMs have been left out of the evaluations. For instance, AI21 Labs Inc.’s Jurassic and Jamba, Cohere Inc.’s Aya and Command LLMs, and xAI's Grok are notably absent from these assessments.
Jupyter AI can automatically generate course learning objectives for over 100 engineering courses based on brief descriptions — GWU Professor Lorena A. Barba showcased in PyCon Colombia 2024. At a keynote titled "Transforming Engineering Education with Jupyter AI," Professor Barba explained that Jupyter AI, using GPT-3.5 and GPT-4 models, can suggest computational activities aligned with the course's learning activities. [Slides Download] While some outputs required refinement, many provided excellent starting points for instructors to build upon. "Perhaps the most exciting part was using 𝙹𝚞𝚙𝚢𝚝𝚎𝚛 𝙰𝙸's "/generate" command to create a raw Jupyter notebook tutorial for a specific course. Through an iterative process of explanation, refinement, and expansion using the AI chat interface, I transformed the notebook into a polished 11-page lab activity on vibration analysis. A task that would typically take perhaps a week was completed in just 2.5 hours with AI assistance," she added. "Jupyter AI is a powerful tool, but it's not a replacement for pedagogical best practices."
Apple didn't pay OpenAI anything for using the GPT-4o model to power tasks on iOS 18, iPadOS 18, and macOS Sequoia, according to a Bloomberg report. Apple seems to have persuaded OpenAI to work for exposure, as the San Francisco-based research lab exposes its technology to hundreds of millions of users. Following this deal, Siri will use ChatGPT capabilities to summarize articles, find photos, or use writing and proofreading tools. Users will only pay more ($20 a month) if they choose a Plus account to access extra perks. In this scenario, Apple will get a commission. Apple is also reportedly in talks with Google to offer Gemini integration to iOS users later this year. Anthropic might offer its Claude AI chatbot as another option. Actually, really liked the Apple Intelligence announcement. It must be a very exciting time at Apple as they layer AI on top of the entire OS. A few of the major themes. Step 1 Multimodal I/O. Enable text/audio/image/video capability, both read and write. These are the native… — Andrej Karpathy (@karpathy) June 10, 2024
