ChatGPT wants to be your shopping portal

OpenAI has recently enabled ChatGPT to search the web for answers, and CEO Sam Altman recently reported that they had provided a billion searches to users in the past week. (For comparison, Google serves over 100 billion searches per week.) But now the other shoe is starting to drop. OpenAI is optimizing its search engine for… scientific research? No. …historical investigation? No. Nothing so tedious. They are optimizing their AI search for … SHOPPING! OpenAI is trying to make it as easy as possible to discover products that you want, to comparison shop different models and vendors, and soon (very soon) they will allow one-button purchasing of the product you select without having to go to the seller’s website. Under the hood, there appears to be some not-fully-disclosed partnering with Shopify. These capabilities are rolling out now to free and paid users alike, so be on the lookout.

ChatGPT optimizes search for shopping

OpenAI’s official announcement on Twitter/X

Space Llama

Facebook/Meta’s open-source AI model, Llama, is going to space. NASA has tasked management consulting company Booz Allen to deploy a customized version of Meta’s model to run on computers aboard the International Space Station. This will allow the crew to access AI capabilities without worries about losing connectivity though a radio link to earth. Since Llama is open source, there are no hidden, black box features to worry about, and the model can be customized to meet the particular needs of the mission.

Llama is going to space

OpenAI CFO says 03-mini is the top competitive coder in the world

All of the major AI companies are racing to optimize their models to help software developers write code. OpenAI has been hinting for a while that it means to win this race, and recently the company’s CFO, Sarah Friar, asserted at a Tech Summit convened by Goldman Sachs that one of OpenAI’s recent models was already the top competitive coder in the world, surpassing all humans. To unpack this statement a bit, there is a website known as Codeforces, that poses coding problems to 2 competitors at once, and the competitor who produces the correct solution first is deemed to be the winner of that match. Competitors are ranked like chessplayers, based on whom they’ve beaten. OpenAI has been trying out its models on Codeforces, and some time ago CEO Sam Altman claimed that one of their models had achieved the 175th best ranking on the site. Now, CFO Friar is claiming that another, later model, called o3-mini, has achieved the top spot. Huge, if true.

OpenAI CFO claims one of their models is better at coding than any human

University of Zurich used AI bots to manipulate users on Reddit

Researchers from the University of Zurich have recently published a paper detailing how they used AI-driven bots to try to influence the opinions of users of online forum Reddit. One of the most popular sub-forums, or “subreddits”, is named “Change My View”, where posters state a position on any topic, and challenge other users to engage in polite debate. If the original poster changes their opinion based on an argument made by another user, the poster awards a “delta” to the persuasive user. The researchers used this setup to test the persuasiveness of AI. First, the AI bot was given a summary of the views and likely demographics of each human poster, as gleaned from their posting history by another AI. Then the AI bot crafted a personalized pitch to change the poster’s views, including false biographical details of the bot itself (“single mother”, hard-working city employee”, etc.). The results were impressive - the authors state that these bots were “surpassing all previously known benchmarks of human persuasiveness.” Not mentioned in this study, which was apparently approved by the university’s review board on human experimentation, was that the study itself was a blatant violation of the terms of service of Reddit, which forbids any undisclosed use of AI in postings.

University of Zurich used AI bots to manipulate users on online forum Reddit

AI helps food banks predict donations and demand

A recent article in the journal Nutrients examines the impact AI is having on food banks. Generally considered a low-tech charitable endeavor, food banks around the world are beginning to take advantage of the publicly available AI tools to help improve their operations. Example projects include using neural networks to improve the accuracy of predictions of supermarket donation availability, optimize collection routes, and reduce transportation costs. Machine learning methods have improved food demand prediction. Sentiment analysis has shed light on donor motivations. To date, these efforts to use AI to support the mission of food banks have been small, local projects, but as some of the initiatives bear fruit, they are likely to spread more widely.

AI helps food banks predict donations and demand

China’s “city brains”

Over 500 cities in China are developing “city brains”, AI-enabled integrated real-time decision support systems for city officials. For example, the city brain project of the Haidian district of Beijing, a municipality of 3 million people, gathers data continuously from over 14,000 CCTV surveillance cameras and over 20,000 environmental sensors. It is used to help plan logistics for impending severe weather, as well as to pre-position police at locations where crime or public protests are expected. Over time, more and more sources of data are planned for inclusion in the city brain, allowing for an ever-broader set of planning and prevention capabilities for the city officials.

China’s “city brains”

Cornell robots learn by watching how-to videos

Training robots to perform tasks has in the past required countless hours of tele-operation, in which the robot is remotely controlled by a human while performing the task until the robot can accomplish it independently. Now researchers at Cornell University have developed an AI framework which allows a robot to learn a task in one shot, simply by watching a how-to video. This is much more similar to how humans learn, and could massively decrease the time and effort required to train robots for any given task. The current system is still experimental, but hopes are high that this can be put into production in the near future.

Cornell robots learn by watching how-to videos

AI enables flying robots to build in the air

Swiss scientists are researching how to use AI to precisely control flying drones to build structures at great heights or in difficult terrain. The aim is to enable repair of tall structures, or to assist disaster recovery, when swarms of semiautonomous drones can bring in emergency shelter materials even when access roads are flooded or destroyed by earthquake. Below is an illustration of the techniques of drone construction that are being studied at the DroneHub at the Swiss Federal Laboratories for Materials Science and Technology, known as EMPA.

AI enables flying robots to build in the air

Children’s Hospital of LA monitors sleep disorders at home with AI

Children with sleep disorders need to be monitored while they sleep in order to get the correct diagnosis, but the need for such services far outstrips the supply of pediatric beds in sleep study centers. Physicians at Children’s Hospital of Los Angeles (CHLA) have come up with a novel solution. They are developing an AI model that can analyze the data collected by an Apple watch on a child sleeping at home in order to diagnose the child’s condition. Apple Watches contain multiple sensors, such as blood oxygen detectors, motion detectors, and more. These data can be correlated by AI to come up with possible diagnoses.

AI analyzes data from Apple Watches to diagnose children’s sleep disorders

AI is revolutionizing nanomedicine

Traditional drug design involves what are called “small molecules”, which fit a certain receptor on cells in our body like a key in a lock. Each molecule of these sorts of traditional drugs is identical to all of the other molecules of that drug. However, many new approaches to therapeutics use biological molecules such as antibodies or DNA, which are produced by living organisms through imprecise biological processes. This branch of therapeutics is called “Nanomedicine”, and AI is now being brought to bear to tame the imprecision and make nanomedicine as predictable as traditional small molecule manufacturing. AI is being used to design novel structures for these biomolecules, in ways that enhance function. One example is called an SNA (for “spherical nucleic acid”), a type of globular DNA which penetrates cells much better than traditional linear DNA, in order to affect cell function to combat a disease.

AI is revolutionizing nanomedicine