Ai Letterhead

AI Frontiers: Industry Shifts, Safety Tools, and Scientific Breakthroughs

September 26, 2024

Ai Letterhead Newsletter

 26 September 2024

Issue # 2

🚀 Top AI Stories This Month 🌐

 

 OpenAI's Chief Technology Officer, Mira Murati, has announced her departure from the company after six and a half years. She mentioned her decision to leave in a memo, expressing her desire to create time for her own exploration. OpenAI is reportedly restructuring to become a for-profit business, though it will retain its non-profit segment. The company is currently pursuing a funding round that could value it at over $150 billion, with Thrive Capital leading the round. However, OpenAI has faced controversy and high-level employee departures, raising concerns among some current and former employees about its rapid growth and safety.

 Microsoft has released new safety, privacy, and security features to help enterprises use generative AI applications. These include Evaluations in Azure AI Studio, Correction, Embedded Content Safety, and Confidential inferencing. These features aim to address challenges such as hallucinations and security risks associated with generative AI. Other tech vendors like AWS and Google have also introduced similar safeguards. Despite these advancements, enterprises still need help implementing these tools and understanding their performance against different models. Microsoft has also learned from past challenges, with new capabilities in Microsoft 365 Copilot addressing transparency in web queries to provide exact web search queries from users' prompts.

 Scientists have been using X-ray crystallography for over 100 years to determine the structure of crystalline materials. MIT chemists have developed a new generative AI model to help determine the structures of powdered crystals, which could aid in characterizing materials for various applications. Freedman and Jure Leskovec, a professor of computer science at Stanford University, are the senior authors of a new study published in the Journal of the American Chemical Society. The study introduces a machine-learning model called Crystalyze, developed to predict structures of powdered crystalline materials using X-ray diffraction patterns. The model, trained on data from the Materials Project, successfully predicted structures for thousands of unsolved X-ray diffraction patterns. It also discovered structures for three materials created in the lab. This approach could lead to the development of new materials with unique properties.

💡 AI Innovation Spotlight 💎

Creating a robust contact list is vital, but manually gathering emails and social profiles is a time-consuming and tedious process. Imagine being able to automatically extract email leads from any website in a matter of seconds.

Meet MyEmailExtractor—your AI-powered solution for fast, effortless lead generation.

Best features

🤖 Automatically extract emails and contact data from websites with ease and precision.

📊 Gather emails, social media profiles, and business information all in one click.

📁 Export all extracted data directly into an XLS or CSV file for efficient lead management.

Designing standout packaging can be a challenge. Many brands face difficulties with complex design processes and the high costs of professional photography, which can limit their creativity and efficiency. What if there was a way to create impressive product packaging and take beautiful product photos without requiring extensive design experience?

Introducing Packify.ai:

Best features

🎨 AI Design: Instantly create customizable packaging—no designer needed!

📸 AI Photoshoot: Generate realistic product images without prototypes, saving time and costs!

Professionals in fast-paced industries such as construction and design constantly seek efficient, mobile tools for seamless collaboration outside of traditional office settings. Imagine having a powerful solution that allows direct and clear visual communication from anywhere.

Introducing PinMy:

Best features

Streamlined Communication: Instant annotations and comments directly on visual content.

Enhanced Accessibility: Accessible across devices, ensuring collaboration from anywhere.

Comprehensive Support: Multi-language and voice-to-text features simplify global teamwork.

AI like ChatGPT is a game-changer, transforming how we work. But often, leveraging AI means constantly switching between tabs or apps, disrupting your workflow.

What if AI seamlessly integrated into your everyday tools, enhancing productivity and creativity across any website or application you use?

Say hello to Flot.ai

Best Features

Works seamlessly across any website and app, integrating effortlessly into your workflow.

Preset expert prompts help you write better and read faster with just one click.

Flot can also help you save memos, screenshots, files, and links with just one click.

📚 Must-Read AI Paper 🔍 

Publications by Autumn Toney-Wails, Christian Schoeberl, James Dunham

Finding scientific publications can be expensive and require expert annotation when researching rapidly advancing fields. There is a lack of widely accepted classification criteria or field taxonomies in areas like artificial intelligence (AI) that cover new topics and technologies. We created a practical definition of AI research based on expert labels to deal with these challenges. We then tested how well the latest chatbot models can annotate expert data. By adjusting the prompts for GPT chatbot models using the arXiv publication database, we developed an alternative automated expert annotation process that achieves 94% accuracy in assigning AI labels. We also fine-tuned SPECTER, a transformer language model pre-trained on scientific publications, which achieved a 96% accuracy in classifying AI publications, only 2% higher than GPT. Results show that chatbots can be dependable data annotators, even in specialized fields, when engineered with effective prompts. To assess the usefulness of chatbot-annotated datasets for future classification tasks, we trained a new classifier on GPT-labeled data and compared its performance to a model trained on arXiv data. The classifier trained on GPT-labeled data outperformed the arXiv-trained model by nine percentage points, achieving 82% accuracy.

 🔗 Further Reading 🧠

Probabilistic forecasting involves estimating the future probability distribution of a time series based on its past data. This is essential for optimizing business processes. For instance, in retail, probabilistic demand forecasts are crucial for ensuring the right inventory is available at the right time and place. This paper introduces DeepAR, a method for generating accurate probabilistic forecasts. It achieves this by training an autoregressive recurrent neural network model on a large number of related time series. The paper demonstrates how applying deep learning techniques to forecasting can address many challenges faced by traditional approaches. Through extensive empirical evaluations of real-world forecasting datasets, the methodology is shown to produce more accurate forecasts than other state-of-the-art methods, with minimal manual work required.

The advancement of artificial intelligence (AI) is expected to significantly impact various aspects of modern life, including transportation, health, science, finance, and the military. In order to adjust public policy accordingly, it is important to have a better understanding and anticipation of these advancements. In a recent survey of machine learning researchers, it was found that they predict AI will surpass human performance in several activities within the next ten years. These activities include translating languages (by 2024), writing high-school level essays (by 2026), driving a truck (by 2027), working in retail (by 2031), writing a bestselling book (by 2049), and working as a surgeon (by 2053). According to the researchers, there is a 50% chance of AI outperforming humans in all tasks within 45 years and automating all human jobs within 120 years. It's worth noting that Asian respondents expect these dates much sooner than those from North America. These findings will be valuable for discussions among researchers and policymakers to help anticipate and manage the impact of AI advancements.

 In our research, we propose a novel deep-learning framework aimed at enhancing the modelling of urban venue popularity and growth. Leveraging a dataset from Foursquare, we focus on modelling individual venues and venue types across London and Paris. Our approach involves representing cities as interconnected networks of venues, allowing us to quantify their structure and identify a robust community framework within these retail networks. Furthermore, we introduce a deep learning architecture that integrates spatial and topological features into a temporal model for predicting venue demand at subsequent time steps. Our experiments demonstrate that our model effectively captures spatio-temporal trends in venue demand, consistently outperforming baseline models. These results underscore the potential of complex network measures and GCNs in constructing predictive models for urban environments, with potential applications in enhancing the modelling of venue demand and growth within the retail sector.

 

Stay curious, stay informed!

 

Ai Letterhead