Daily AI Newsletter: OpenAI new reasoning technology, code name ‘Strawberry’

In today’s email:

🦷 AI system achieves 96% accuracy in determining sex from dental X-rays
🕵🏻‍♂️ Experiment finds AI boosts creativity individually — but lowers it collectively
🇺🇸 U.S. Rep. Jennifer Wexton debuted an AI-generated version of her voice.
🧰 8 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

OpenAI working on new reasoning technology under code name ‘Strawberry’

OpenAI, the creator of ChatGPT, is working on a new project code-named "Strawberry" to enhance the reasoning capabilities of its AI models, according to internal documents reviewed by Reuters. The project, which remains a closely guarded secret, aims to enable AI to perform deep research by navigating the internet autonomously and planning ahead to tackle complex problems. This development is part of OpenAI's broader strategy to improve its models' reasoning abilities, a critical factor in achieving human-like or super-human intelligence.

The Strawberry project, previously known as Q*, involves a specialized post-training process that fine-tunes pre-trained AI models to hone their performance in specific areas. This approach is similar to the "Self-Taught Reasoner" (STaR) method developed at Stanford, which allows AI models to iteratively create their own training data to boost their intelligence. OpenAI's goal is to use Strawberry to enable its models to handle long-horizon tasks, which require planning and executing a series of actions over extended periods.

OpenAI has been signaling to developers and external parties that it is close to releasing technology with significantly advanced reasoning capabilities. The company aims to use Strawberry to test the AI's ability to conduct autonomous web research and potentially perform tasks typically done by software and machine learning engineers. While the exact details of Strawberry's workings and its release timeline remain undisclosed, this project represents a significant step towards improving AI's reasoning and problem-solving skills.

AI system achieves 96% accuracy in determining sex from dental X-rays

Researchers in Brazil have developed a machine-learning system capable of determining an individual's sex from panoramic dental X-rays with 96% accuracy when the image resolution is high and the individual is older than 16. The study, published in the Journal of Forensic Sciences, highlights the importance of forensic dentistry, particularly in scenarios where only the jaw and teeth are available for analysis, such as in mass accidents or advanced decomposition cases. The AI-powered tool utilizes convolutional and residual networks, showing that AI can assist in faster and more accurate interpretations of dental radiographs.

The study involved analyzing 207,946 panoramic radiographs from 15 clinical centers in São Paulo, Brazil, with 58% of the images from female patients. The researchers trained two machine-learning algorithms, both achieving similar accuracy levels. They found that the resolution of the images and the age of the individuals were critical factors affecting accuracy. For patients aged 20 to 50, the system's accuracy was over 97%, slightly lower for older patients, and significantly lower for younger children, with only 74% accuracy for those under six years old.

Despite the promising results, the researchers noted that all images used in the study were from living individuals, suggesting that the effectiveness might differ with images from decomposed human remains. The study underscores the potential of AI in forensic science, offering a new tool for sex determination from panoramic radiographs and emphasizing the importance of image quality and patient age in achieving high accuracy.

Experiment finds AI boosts creativity individually — but lowers it collectively

A new study by Anil Doshi and Oliver Hauser, published in Science Advances, explores the impact of AI on creative tasks, specifically in writing short stories. The research found that while AI assistance, such as using GPT-4, boosted creativity in individuals who were less naturally creative, it collectively lowered the creativity of the group. This mixed result suggests that while AI tools can be helpful for some, they might not offer truly novel ideas in creative endeavors, echoing a sentiment shared by many in the creative community.

The experiment involved hundreds of participants writing short stories, with groups varying in their access to AI-generated story ideas. One group wrote without assistance, another could consult GPT-4 for a single idea, and a third could get up to five suggestions. The stories were then evaluated on novelty, usefulness, and emotional enjoyment by both their authors and a separate group unaware of the AI assistance. The findings indicated that less creative individuals benefited significantly from AI, while those with higher creativity saw no improvement.

Hauser emphasized the importance of rigorously evaluating AI's impact on creativity rather than assuming its benefits. The study used a word-production task to approximate participants' creativity levels, acknowledging the complexity and debates around measuring creativity. This research highlights the nuanced effects of AI on creativity, suggesting that while AI has potential, it must be carefully integrated into creative processes.

Amazon’s AI chatbot Rufus is now live for all US customers

Earlier this year, Amazon introduced Rufus, a generative AI-powered shopping assistant designed to help customers make more informed purchase decisions by answering a variety of shopping-related questions. Now available to all U.S. customers in the Amazon Shopping app, Rufus acts as a virtual shopping assistant, providing information on product details, customer reviews, and community Q&As. Customers can ask specific questions like “Is this coffee maker easy to clean?” and receive detailed answers. Rufus also surfaces related questions and highlights summaries of customer reviews, making it easier for shoppers to understand product features and customer feedback.

Rufus has been particularly helpful in offering product recommendations and comparisons. Customers have found it useful for selecting items with specific features, such as a pool umbrella suitable for Florida’s weather or the best wireless outdoor speakers. Rufus can compare products like gas vs. wood-fired pizza ovens or OLED vs. QLED TVs, helping customers make well-informed decisions. It also keeps users updated on the latest products and trends, answering questions about advanced Fire tablets for kids or the newest denim styles for women.

In addition to shopping-related inquiries, Rufus assists with accessing current and past orders, tracking packages, and providing information based on previous purchases. Customers can quickly find out when their orders will arrive or when they last ordered a specific item. Rufus also answers broader questions, such as what’s needed to make a soufflé or prepare for a summer party, showcasing its versatility. As Amazon continues to improve Rufus, the company looks forward to seeing how customers utilize this tool to enhance their shopping experience.

Other stuff

What exactly is an AI agent? 🔥
AI Has Become a Technology of Faith
Universities Don’t Want AI Research to Leave Them Behind
What Large Language Models Can Or Can't Do 🔥
Can ChatGPT do data science?
Gen-3 Alpha can simulate liquids such as water, paint, oil, honey and molten glass. 🔥
U.S. Rep. Jennifer Wexton debuted an AI-generated version of her voice. 🔥
Collection of more than 70,000 public domain books that can now not only be read, but also listened to, using the latest Text-to-Speech model from OpenAI. 🔥