Today’s top AI Highlights:
Apple will turn to AI processing in the “cloud black box”
Scale AI releases SEAL Leaderboards: A new way to evaluate leading AI models
OpenAI releases ChatGPT versions designed for universities and non-profit organizations
Transform search into visually stunning content with Perplexity AI’s new tool
& so much more!
Read time: 3 mins
Latest Developments 🌍
Secure Cloud AI with Black Box Processing 🔏
Apple has been pushing on-device AI for years, but now they’re venturing into the cloud. This raises privacy concerns, right? Well, they might have got a solution. They’re planning to use “confidential computing” techniques to keep user data private even while processing it on their servers. This “black box processing” approach keeps data encrypted throughout the entire process, meaning not even Apple can access it.
Key Highlights:
Apple has been working on confidential computing for at least three years. This is not a last-minute solution but a carefully crafted approach to balancing privacy with powerful AI features.
The system is designed to be so secure that Apple won’t even be able to provide user data in response to subpoenas or government requests. This is a significant step forward in data privacy, especially in an era of increasing government surveillance.
Apple is planning to use this technology to create lightweight wearables that offload processing to the cloud. This could lead to a new generation of devices with smaller batteries and more advanced features.
Private and Expert-Driven Evaluations by Scale AI 🤫
LLMs are constantly surprising us with their abilities but their true reasoning abilities remain a point of debate. Some models may be excelling on benchmarks simply due to “overfitting”. They’ve memorized data from those benchmarks during training, rather than demonstrating genuine understanding. To address this issue and provide a more trustworthy evaluation process, Scale AI has released SEAL Leaderboards. These leaderboards offer a private and secure way to assess leading AI models, including GPT-4, Gemini, Claude, Mistral, and others.
Key Highlights:
Comprehensive Evaluations: SEAL Leaderboards assess models on coding, math, instruction following, and multilingual capabilities (Spanish).
Reducing Overfitting Risk: In their math evaluations, Scale AI used the GSM1k benchmark, which was specifically designed to mirror the popular GSM8k benchmark but without the risk of data contamination. Their findings suggest that some models show significant performance drops when evaluated on GSM1k, indicating potential overfitting to the original GSM8k.
Private and Unexploitable: SEAL Leaderboards use private evaluation datasets that are kept hidden from the models, ensuring that models cannot overfit to the specific evaluation data. This leads to a more objective and accurate measure of their capabilities.
Expert-Driven and Continuously Updated: SEAL Leaderboards rely on evaluations conducted by domain experts, guaranteeing high quality and credibility. The leaderboards are regularly updated with new datasets and models to reflect the rapidly evolving field of AI.
Quick Updates from OpenAI 🤌
Starting today, all ChatGPT free users can now use browse, vision, data analysis, file uploads, and GPTs within ChatGPT.
OpenAI has announced ChatGPT Edu, a ChatGPT version powered by GPT-4o, built for universities to leverage AI for education. Besides the functionalities available in ChatGPT free tier, it also includes enterprise-level security and controls, gives the ability to create and share custom GPTs, and also has higher rate limits.
The company has announced OpenAI for Nonprofits initiative with which nonprofit organizations can now access ChatGPT Team at a discounted rate of $20 per month per user.
OpenAI is expanding its umbrella of partnerships with news publications. It has now partnered with The Atlantic and Vox Media group. These media houses will have “privileged” access to OpenAI’s technology, and in turn their articles will be discoverable in ChatGPT.
✍️ Build Personalized Marketing Chatbots with Google Gemini
Learn how to build personalized marketing chatbots with Google Gemini and LoRA in just 60 minutes. Join this FREE webinar for live demos and hands-on code sharing. Register now before the spots get filled!
😍 Enjoying so far, share it with your friends!
Tools of the Trade ⚒️
Perplexity Pages: Turn your research into visually appealing, comprehensive articles with AI-generated and custom visuals. You can create, organize, and share well-structured articles on any topic, tailoring them to your audience.
Clearspace: Earn your screen time with exercise. It uses AI to detect and count pushups to unlock your apps based on physical activity. By enforcing a time buffer and requiring exercises like pushups to earn screen time, it helps you reduce digital distractions and form healthier phone habits.
Zanki: A phonics-based reading app that uses AI-powered flashcards and spaced repetition to help children learn to read efficiently. It adapts to your child’s reading level and interests, rewarding them with customizable stickers for motivation.
Awesome LLM Apps: Build awesome LLM apps using RAG for interacting with data sources like GitHub, Gmail, PDFs, and YouTube videos through simple texts. These apps will let you retrieve information, engage in chat, and extract insights directly from content on these platforms.
Hot Takes 🔥
In the near future, the main purpose of most human beings will be to provide training data for AI models. ~
Bindu ReddyApple outsourcing its chatbot to OpenAI is like IBM outsourcing its PC operating system to Microsoft. ~
Pedro Domingos
Meme of the Day 🤡
How paying for ChatGPT subscription feels like now
That’s all for today! See you tomorrow with more such AI-filled content.
Real-time AI Updates 🚨
⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!
PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!