The vast number of festive socks you got over the past several years’ holidays bears testament to the global lack of imagination when it comes to gifts. We can do better, and we will, with the help of OpenAI’s GPT-3 .

More specifically, I urge you to try out GiftGenius , a handy AI concierge that claims to use the GPT-3 language model to recommend suitable gifts for everyone in your life.

Simply describe a recipient with something like ‘my buddy from college who is 35, enjoys photography, and has a three-year-old son,’ and the bot will parse your query to come up with surprisingly decent suggestions.

In this case, it spat out recommendations for camera-equipped drones for my friend to capture aerial photos and videos of his son, as well as DSLR lenses “to get better shots at sporting events or on family adventures,” and “A photo book of his sons first year To help him relive all the memories he has missed while at work or traveling for business (sic).”

You can adjust a slider beneath your query to receive less creative or more creative suggestions, and the items are all available on Amazon US. The site also breaks these up into sections, each corresponding to ideas GPT-3 came up with based on your query.

GiftGenius is a clever implementation of OpenAI‘s powerful language model, and it generally does a good job of coming up with suggestions because the working premise makes sense. When you describe the person you’re shopping for, you mention things about who they are, and that allows for various prompts for gifts they might enjoy.

The site came up with a bunch of neat ideas I may not have otherwise thought of for other people on my list: an aromatherapy oil diffuser for a friend who’s into meditation, a gardening journal for my mother who’s currently obsessed with house plants.

Even the ‘less creative’ options are useful (like a compost bin and a DIY automatic irrigation kit for my gardener mom), because they surface items you might not have spotted with a straightforward keyword search directly on Amazon or a similar web store.

Occasionally, some of the recommended items may be off by a little or a lot (my aforementioned query brought up a set of shot glasses, like because of GPT-3’s word association from ‘photography’ to ‘shots’). But hey, at least GiftGenius isn’t suggesting goddamn socks. Am I glad I came across it just in time for the holidays.

Just so you know, the project makes money when you click through to items listed on the site, as the URLs appear to have an affiliate code attached. We’ve contacted the developers to learn more about how GiftGenius works and if they have plans to develop it further, and we’ll update this story when we hear back.

Google’s new AI automatically turns webpages into videos

If you have a product webpage, you’d wish that you could create a handy video to show to clients with all the info. However, you might not have the budget to hire someone to make it. To solve that, Google’s AI team is working on a future solution that automatically converts webpages into videos.

Google’s URL2Video tool helps you convert your website into a short video if you specify the constraints of the output video, such as the duration and aspect ratio. The tool tries to maintain the design language of the source page and uses its elements such as the text, images, and clips to create a new video.

To train the model, the company interviewed designers to determine important aspects of a webpage derived from heuristics. Based on these parameters, the tool analyzes a page and ranks key elements. Then based on conditions provided by the users, it selects top elements such as headers and images to churn out a video.

Once the video is generated, you can change colors and style to regenerate the footage according to your needs. You can see an example of the tool being used to create a video for Google Search below.

Google’s not the only company working on this. Baidu has built and rolled out an AI in the limited capacity that can generate news videos with voice over using a single URL.

In the future, Google’s AI team wants to add the ability to add audio tracks and voiceovers during the editing process.

You can read about the tool’s technical details in this paper.

This AI can tell if you have prostate cancer by looking at your pee

Researchers at the Korean Institute of Science and Technology (KIST) recently developed a machine for detecting prostate cancer that only needs 20 minutes of your time and a few ounces of your pee to achieve near 100% accuracy. Human oncologists are only about 30% accurate when it comes to detecting the disease. This is a big deal.

Background: Detecting prostate cancer is, quite literally, a pain in the ass. Under the current paradigm the disease is confirmed through a combination of lab work and invasive diagnostics. This involves a painful biopsy procedure where surgeons remove a tissue sample from the prostate gland itself.

Unfortunately a large number of patients who endure this procedure don’t actually need it. These otherwise healthy people risk hospital infection, surgical death, and lingering side-effects including discomfort, pain, and internal bleeding.

How it works: The KIST team decided to focus on urine because it contains trace amounts of what researchers refer to as “cancer factors.” Typically, humans can’t diagnose prostate cancer using urine because the concentration of these cancer factors is simply not high enough to withstand the standard testing methods.

To overcome this obstacle, the team used a special semiconductor-based sensor sensitive enough to detect enough data for the team’s algorithms to parse and correlate.

Per a press release from the Korean National Council of Science and Technology:

Quick take: Wow! This is awesome. Assuming everything in the research pans out when scaled to the general population, this could save a lot of lives. On average about 1 in every 41 men will die of prostate cancer, it’s the second leading cause of cancer-related death among men worldwide.

Best of all, the team believes this work can be adapted for other types of cancer.

You can read the team’s research paper here .