Keeping up with an industry as fast-paced as artificial intelligence is a tall order. So, until an AI can do it for you, here’s a helpful roundup of recent stories in the world of machine learning, along with notable research and experiments we didn’t cover on our own.
This week, Amazon announced Rufus, an artificial intelligence shopping assistant trained on the e-commerce giant’s product catalog as well as information from around the web. Rufus lives inside the Amazon mobile app, helping you find products, make product comparisons, and get recommendations on what to buy.
From extensive research at the start of a shopping trip, such as “what to consider when buying running shoes?” in comparisons like “what are the differences between trail and street running shoes?” … Rufus fundamentally improves how easy it is for customers to find and discover the best products to meet their needs,” Amazon writes in a blog post.
This is all great. But my question is who is calling it Really?
I’m not convinced that GenAI, particularly in chatbot form, is a piece of technology that the average person cares about — or even thinks about. Research backs me up on this. Last August, the Pew Research Center found that among those in the US who have heard of OpenAI’s GenAI ChatGPT (18% of adults), only 26% have tried it. Of course, use varies by age, with a higher proportion of young people (under 50) reporting use than older people. But the fact remains that the vast majority don’t know—or don’t care—to use what is arguably the most popular GenAI product out there.
GenAI has its well-publicized problems, including a tendency to fabricate facts, infringe copyright, and emit bias and toxicity. Amazon’s previous attempt at a GenAI chatbot, Amazon Q, struggled mightily – leaking confidential information on its first day of launch. But I’d argue that the biggest problem with GenAI right now—at least from a consumer perspective—is that there are few generally compelling reasons to use it.
Sure, GenAI like Rufus can help with specific, narrow tasks like shopping by occasion (e.g. finding winter clothes), comparing product categories (e.g. the difference between lip gloss and oil), and looking top deals (eg Valentine’s Day gifts). However, does it meet the needs of most buyers? Not according to recent voting by e-commerce software startup Namogoo.
Namogoo, which asked hundreds of consumers about their online shopping needs and frustrations, found that product images were by far the most important factor in a good e-commerce experience, followed by product reviews and descriptions. Respondents ranked search as the fourth most important and “easy navigation” as the fifth. Remembering preferences, information and purchase history was last.
The bottom line is that people generally shop with a product in mind. that search is an afterthought. Maybe Rufus will shake up the equation. I’m inclined to think not, especially if it’s a difficult release (and it may well be given the reception from Amazon’s other GenAI shopping experiments) — but I guess stranger things have happened.
Here are some other notable AI stories from the past few days:
- Google Maps is experimenting with GenAI: Google Maps introduces a GenAI feature to help you discover new places. Leveraging large language models (LLM), the feature analyzes the more than 250 million locations on Google Maps and contributions from more than 300 million Local Guides to generate recommendations based on what you’re looking for.
- GenAI tools for music and more: In other Google news, the tech giant released GenAI tools for creating music, lyrics and images, and brought Gemini Pro, one of its most capable LLMs, to Bard chatbot users worldwide.
- New open AI models: The Allen Institute for AI, the nonprofit AI research institute founded by the late Microsoft co-founder Paul Allen, has released several GenAI language models that it claims are more “open” than others — and, crucially, licensed in such a way that developers can use unlimited for training, experimentation and even commercialization.
- FCC moves to ban AI-generated calls: The FCC is proposing to outlaw the use of voice-cloning technology in robocalls, making it easier to charge the operators of these scams.
- Shopify Releases Image Editor: Shopify releases a GenAI media editor to improve product images. Marketers can choose a type from seven styles or type in a question to create a new background.
- GPT, referred to: OpenAI pushes adoption of GPTs, third-party apps powered by its AI models, by enabling ChatGPT users to refer to them in any conversation. Paid ChatGPT users can bring GPTs into a chat by typing “@” and selecting a GPT from the list.
- OpenAI works with Common Sense: In an unrelated announcement, OpenAI said it is partnering with Common Sense Media, the nonprofit organization that reviews and ranks the appropriateness of various media and technology for children, to collaborate on AI guidelines and educational materials for parents, educators and young adults .
- Autonomous browsing: The Browser Company, which makes Arc Browser, is trying to build an artificial intelligence that surfs the web for you and gives you results by bypassing search engines, writes Ivan.
More machine learning
Does an AI know what is “normal” or “typical” for a given situation, medium, or expression? In some ways, large language models are uniquely suited to identifying patterns that are most similar to other patterns in their datasets. And indeed that’s what Yale researchers found in their research into whether an artificial intelligence could recognize the “typicality” of one thing in a group of others. For example, given 100 romance novels, which is the most and which is the least “typical” given what the model has in store for that genre?
Interestingly (and disappointingly), Professors Balázs Kovács and Gaël Le Mens worked for years on their own model, a variant of BERT, and just as they were about to publish, ChatGPT came out and in many ways copies exactly what they did. “You could cry,” Le Mens said in a press release. But the good news is that the new AI and their old, tuned model suggest that indeed, this type of system can determine what is typical and atypical in a data set, a finding that could be useful down the line. Both point out that although ChatGPT supports their thesis in practice, its closed nature makes scientific work difficult.
Scientists at the University of Pennsylvania were examining another strange concept to quantify: common sense. Asking thousands of people to rate statements, things like “you get what you give” or “don’t eat food past its expiration date” on how “social” they were. Not surprisingly, although patterns emerged, “few beliefs were recognized at the group level.”
“Our findings suggest that each person’s idea of common sense may be uniquely their own, making the concept less common than one might expect,” says co-lead author Mark Whiting. Why does this appear in an AI newsletter? Because like almost everything else, it turns out that something as “simple” as common sense, which one would expect AI to eventually acquire, is not simple at all! But by quantifying it this way, researchers and testers may be able to tell how much common sense an AI has, or which groups and biases it aligns with.
Speaking of biases, many large language models are pretty loose with the information they take in, meaning that if you give them the right prompt, they can respond in ways that are offensive, incorrect, or both. Latimer is a startup that aims to change that with a model designed to be more inclusive.
While there aren’t many details about their approach, Latimer says their model uses Retrieval Augmented Generation (he thinks it will improve the answers) and a bunch of uniquely licensed content and data that comes from many cultures not typically represented in these databases. So when you ask about something, the model doesn’t go back to some 19th century monograph to answer you. We’ll know more about the model when Latimer releases more information.
Image Credits: Purdue / Bedrich Benes
However, one thing an AI model can certainly do is grow trees. Fake trees. Researchers at Purdue’s Digital Forestry Institute (where I’d like to work, call me) have built an extremely compact model that simulates the growth of a tree realistically. This is one of those problems that looks simple but isn’t. You can simulate tree growth in action if you’re making a game or movie, sure, but what about serious scientific work? “Although artificial intelligence has become seemingly pervasive, it has so far proven very successful in modeling 3D geometries unrelated to nature,” said lead author Bedrich Benes.
Their new model is about one megabyte, which is extremely small for an AI system. But of course DNA is even smaller and denser, and encodes the entire tree, from root to bud. The model still works in the abstract—it’s by no means a perfect simulation of nature—but it shows that the complexities of tree growth can be encoded in a relatively simple model.
Latest, a robot from Cambridge University researchers that can read braille faster than a human, with 90% accuracy. Why do you ask? In fact, it’s not for the blind to use – the team decided this was an interesting and easily quantified task to test the sensitivity and speed of robotic finger tips. If he can read braille just by zooming in on it, that’s a good sign! You can read more about this interesting approach here. Or watch the video below:
