TikTok launched a text-to-image AI that generates abstract backgrounds from prompts. Clean Robotics raised $4.5 million for Trashbot, a recycling bin with 90% sorting accuracy. A Midjourney image trained on Swedish artist Simon Stålenhag's style sparked backlash over AI plagiarism.
Chapters
Hi there everyone. Welcome to another episode of the weekly roundup. This week we're talking about a robot that can sort recyclables, TikTok's image generating AI, and how data science could save the lives of firefighters in the future. Don't forget to let us know which stories you're finding most interesting this week and make sure you're subscribed so that you never miss another update on the latest and greatest in data science. For our first story this week, Clean Robotics have announced a $4.5 million successful Series A funding round that will go towards Prolifera, the company's recycling tech. The product spearheading the company's growth is their Trashbot, a bin that can sort recyclable materials from waste to correct human error and improve the quality of recycled materials. Trashbot uses onboard machine learning and robotics systems to sort materials from a single disposal point, and Clean Robotics claims that the machines can do so with around 90% accuracy. The company's CEO Charles Yap said in a release, recycling rules are confusing and consumers are often so confused that their recycling accuracy is less than chance, leading to highly contaminated recyclables. Our system improves material diversion from landfills, resulting in more recyclables and less waste. Given the onboard AI, the trash sorting robot is gathering data to help improve the sorting process. Trashbot also generates data for waste audits, triggers fullness alerts, and features a large display for video content, making it an attractive investment for high traffic areas such as malls, airports and pedestrian thoroughfares. But do you ever use an automatic rubbish sorting system? Let us know in the comments. In our next story, TikTok has begun to offer users a text to image AI that generates an abstracted background image based on a natural language prompt. Compared to some of the bigger players in the AI image generating space, TikTok's text to Image AI is limited. It creates only rather abstract and swirling images. Compared to other models that can produce photorealistic imagery and coherent illustrations that looked like they were drawn or painted by humans, text to image AI still requires a lot of processing power, and considering TikTok's large user base, providing users with the ability to create photorealistic imagery for free on smartphones would be incredibly resource intensive. Some have also argued that the level of abstraction prevents the AI from being misused to create harmful imagery, a concern with any widely available image creation software. The AI green screen effect certainly shows how quickly this technology is going mainstream, and we're looking forward to seeing all the weird and wonderful creations come up with in the app. With the ongoing proliferation of image generating AI software, it should come as no surprise that questions are being raised about whether such technology will replace artists and illustrators. For our next story, we'll take a look at how AI mimicked a Swedish artist style and how the use of AI generated images in newspapers has sparked a heated debate. Simon Stalenhag, a Swedish artist known for haunting paintings that blend nature and machinery, is a vocal critic of AI generated images. Naturally, when a reader in Intellectual Property Law at the University of Sussex fed the artist's style into the midjourney Image generator and then posted AI generated images in Stalinhag's style on Twitter, the artist was pretty upset. After a lengthy and heated debate on Twitter, the artist acknowledged that while he doesn't view the AI images as plagiarism, he strongly disapproves of his style being used to enrich powerful companies and CEOs. Similarly, when another midjourney generated image was used as the COVID image for Charlie Warzel's Galaxy Brain newsletter in the Atlantic, artists called the writer out for not commissioning an illustrator for the image. The tweet calling Wozel out quickly went viral, and while he explained that the newsletter was never a candidate for getting a sonar by the Atlantic's art department, the Twitter war raged on. Definitely take a look at Charlie Wozel's article I Went Viral in a Bad Way, which we'll link in the description below and let us know in the comments what you think about this fascinating issue facing AI imagery. In our final story this week, a new AI model might be able to save the lives of firefighters in the future by providing them with more accurate estimates of when a flashover event is likely to happen. In firefighting, a flashover event is when almost all the combustible items in a room suddenly ignite and typically burn at around 600 degrees Celsius. Predicting these moments is difficult, thanks to the different layout and contents of each building, and current methods of predicting flashovers rely on constant streams of temperature data from sensors within the building. Machine learning already assists in filling the gaps in information when sensors fail, but researchers at the National Institute of Standards and Technology and the Hong Kong Polytechnic University have developed a system called FlashNet that aims to rethink temperature modeling and flashover predictions to cope with the variability of real fires. The researchers beefed up their approach with graph neural networks, a kind of machine learning algorithm good at making judgments based on graphs of nodes and lines representing different data points and their relationships with one another. These algorithms are commonly used to predict travel times. This new way of thinking about fires resulted in an accuracy of at best 92.1% with 30 seconds of lead time and outperformed five other machine learning based models, including the author's previous model. Critically, the tool produced the least false negatives dangerous cases where the models fail to predict an imminent flashover. The team plans to build and burn their own versions of buildings in order to test the model for use in the real world. That brings us to the end of another roundup. Thanks for watching. We hope you enjoyed this week's episode and let us know in the comments which articles you're finding most interesting this week. Week. We'll see you same time, same place next week for your weekly dose of the latest and greatest in data science.
