Staying Informed in the Rapidly Evolving AI Landscape
Navigating the fast-paced world of artificial intelligence (AI) can be challenging. Until there's AI that can do it for you, here’s a concise roundup of the latest developments in machine learning, including notable research and experiments we haven’t covered previously.
This week, Amazon unveiled Rufus, a shopping assistant powered by AI, designed to enhance the e-commerce experience. Trained on Amazon’s extensive product catalog and web information, Rufus is integrated into the Amazon mobile app to help users find products, compare items, and receive tailored recommendations.
“From broad queries at the start of a shopping journey, like ‘what should I consider when buying running shoes?’ to comparisons such as ‘what are the key differences between trail and road running shoes?’ . . . Rufus significantly streamlines how customers discover the best products that meet their requirements,” Amazon explained in a recent blog post.
While this sounds promising, the real question is: who actually wants it?
I remain skeptical that generative AI, especially in chatbot form, resonates with the everyday consumer. Survey data supports my viewpoint. Last August, the Pew Research Center reported that of U.S. adults who were aware of OpenAI’s ChatGPT, only 26% had actually tried it. Younger individuals (under 50) are more likely to engage with the technology, but it's clear that most people are either unaware or uninterested in using one of the most well-known generative AI products available.
Generative AI faces significant challenges, including the propensity to fabricate information, copyright infringements, and biases. Amazon's earlier effort with a generative AI chatbot, Amazon Q, faced serious issues, including the premature disclosure of sensitive information. From a consumer standpoint, the underlying problem with generative AI is the lack of compelling reasons for its widespread use.
Certainly, tools like Rufus can assist with specific shopping tasks—whether it’s finding winter apparel, comparing product categories, or suggesting gifts for special occasions. However, does it genuinely cater to most shoppers' needs? According to a recent survey by e-commerce software startup Namogoo, insights suggest otherwise.
Namogoo surveyed hundreds of consumers about their online shopping experiences and found that product images were the primary element contributing to a positive e-commerce experience, followed closely by product reviews and descriptions. Search functionality ranked fourth, while "simple navigation" was fifth; personal shopping history and preferences were deemed less important.
These findings imply that consumers typically have specific products in mind when shopping, treating search as a secondary consideration. While Rufus may introduce some changes, I remain doubtful it will significantly alter consumer behavior—especially given previous challenges Amazon has faced with its AI shopping initiatives.
Here are some other noteworthy AI developments from recent days:
- Google Maps' Generative AI Feature: Google Maps has launched a generative AI feature designed to help users discover new locations. Utilizing large language models (LLMs), this feature examines over 250 million sites on Google Maps, incorporating input from more than 300 million Local Guides to provide tailored suggestions.
- New AI Tools for Music Creation: In other news, Google has rolled out generative AI tools that allow users to create music, lyric compositions, and images. Additionally, the advanced Gemini Pro LLM has been made available to Bard chatbot users worldwide.
- Open AI Models Release: The Allen Institute for AI, founded by the late Microsoft co-founder Paul Allen, has introduced new generative AI language models, described as more "open" and freely licensed for developers to use for training, experimentation, and commercialization.
- FCC's Response to AI-generated Calls: The FCC is considering making it illegal to use voice-cloning technology in robocalls, simplifying the process of holding fraudsters accountable.
- Shopify's Image Editing Tool: Shopify has launched a generative AI media editor, enabling merchants to enhance product images by either choosing from seven styles or entering a custom prompt for background generation.
- OpenAI's GPTs Integration: OpenAI is promoting the use of third-party apps powered by its AI models, allowing ChatGPT users to include these apps in their conversations by typing “@” and selecting from a list.
- Collaboration with Common Sense Media: OpenAI announced a partnership with Common Sense Media, a nonprofit focused on reviewing the suitability of media and tech for young people, to develop AI guidelines and educational resources for parents and educators.
- Autonomous Browsing Development: The Browser Company is working on an AI that autonomously navigates the web to deliver results, circumventing traditional search engines.
Exploring Further into Machine Learning
Research at Yale has questioned whether AI can recognize what constitutes "normal" or "typical" in various scenarios. Through their study, researchers Balázs Kovács and Gaël Le Mens explored whether AI can identify typicality among items in a dataset, such as discerning the most and least typical romance novels based on existing genre data.
While working on their model inspired by BERT, they found themselves overshadowed by the emergence of ChatGPT, which had recreated much of their research. Despite this setback, both their model and ChatGPT's functionalities indicate that AI can indeed differentiate typical from atypical elements within datasets—an insight that might prove valuable in future applications, although the closed source nature of models like ChatGPT complicates scientific inquiry.
Another intriguing study from the University of Pennsylvania aimed to quantify the concept of common sense. By having thousands of participants evaluate phrases such as “you get what you give,” researchers discovered that individual interpretations of common sense can vary markedly. As co-lead author Mark Whiting notes, “Our findings suggest that each person’s idea of common sense may be uniquely their own.” This has important implications for AI development, as it reveals how subjective notions like common sense are complex and multifaceted.
On the topic of biases, many large language models are known for their loose information handling, which can lead to questionable outputs based on user prompts. Startups like Latimer are addressing this issue by developing models designed to be more inclusive at their core, leveraging retrieval-augmented generation and diverse licensed datasets.
In a fascinating study at Purdue University, researchers developed a compact AI model that realistically simulates tree growth—an area where traditional simulations have often fallen short. The model, only about one megabyte in size, demonstrates how complex growth patterns can be captured using simplified algorithms, paving the way for future applications.
Finally, researchers from Cambridge University have created a robot capable of reading Braille with 90% accuracy—though it's not intended for assisting visually impaired individuals. This project aimed to evaluate the sensitivity and efficiency of robotic fingertips, showcasing how AI can successfully undertake specific tasks.
For more insights and developments in AI, stay tuned!
Ranking
