Google Deepmind presents Self-Discover.
Large language models self-compose reasoning structures.
Join the discussion on this paper page.
Category: robotics/AI – Page 635
When Lab-Trained AI Meets the Real World, ‘Mistakes can Happen’
Tissue contamination distracts AI models from making accurate real-world diagnoses. Human pathologists are extensively trained to detect when tissue samples from one patient mistakenly end up on another patient’s microscope slides (a problem known as tissue contamination). But such contamination can easily confuse artificial intelligence (AI) models, which are often trained in pristine, simulated environments, reports a new Northwestern Medicine study.
“We train AIs to tell ‘A’ versus ‘B’ in a very clean, artificial environment, but, in real life, the AI will see a variety of materials that it hasn’t trained on. When it does, mistakes can happen,” said corresponding author Dr. Jeffery Goldstein, director of perinatal pathology and an assistant professor of perinatal pathology and autopsy at Northwestern University Feinberg School of Medicine.
“Our findings serve as a reminder that AI that works incredibly well in the lab may fall on its face in the real world. Patients should continue to expect that a human expert is the final decider on diagnoses made on biopsies and other tissue samples. Pathologists fear — and AI companies hope — that the computers are coming for our jobs. Not yet.”
Guided Energy helps EV fleet managers optimize battery charging
Imagine you work for a car rental agency or a package delivery company and you’re in charge of a fleet of vehicles. If you’re switching to EV vehicles, it becomes more complex to manage your vehicles due to long charging time and limited charging point availabilities.
Guided Energy, a French startup that raised $5.2 million from Sequoia Capital and Dynamo Ventures at the end of 2023, is building a software tool that help EV fleet operators when it comes to charge management and dispatch. The company aggregates data from vehicles, public and private charging points and uses machine learning to tell you when and where you’re supposed to charge your vehicles.
“The beauty of the EV ecosystem is that it is all online. This means, we connect to both EVs and charging points directly. Where customers already have telematics or supervision platforms in place, we can integrate with them using APIs into our platform, giving them a single, real-time, unified view of their EV operations,” co-founder and CEO Anant Kapoor told me.
Meta’s Plans to Label AI-Generated Content Are a Sad Fart
Meta is promising to roll out auto-labeling for AI-generated images — as soon as it figures out how, that is.
Nick Clegg, Meta’s president of global affairs, said in a policy update that the company is currently working with “industry partners” to formulate criteria that will help identify AI content. Once those criteria are determined, Meta will begin automatically labeling posts featuring any AI-generated images, video, or audio “in the coming months.”
“This approach represents the cutting edge of what’s technically possible right now. But it’s not yet possible to identify all AI-generated content, and there are ways that people can strip out invisible markers,” Clegg wrote. “So we’re pursuing a range of options. We’re working hard to develop classifiers that can help us to automatically detect AI-generated content, even if the content lacks invisible markers.”
Building a DNA nanoparticle to be both carrier and medicine
Scientists have been making nanoparticles out of DNA strands for two decades, manipulating the bonds that maintain DNA’s double-helical shape to sculpt self-assembling structures that could someday have jaw-dropping medical applications.
The study of DNA nanoparticles, however, has focused mostly on their architecture, turning the genetic code of life into components for fabricating minuscule robots. A pair of Iowa State University researchers in the genetics, development, and cell biology department—professor Eric Henderson and recent doctoral graduate Chang-Yong Oh—hope to change that by showing nanoscale materials made of DNA can convey their built-in genetic instructions.
“So far, most people have been exploring DNA nanoparticles from an engineering perspective. Little attention has been paid to the information held in those DNA strands,” Oh said.
Deep learning-aided decision support for diagnosis of skin disease across skin tones
Deep learning #AI for skin lesions assessed for assistance to 800 dermatologists and primary care physicians from 39 countries Marked improvement in accuracy but widened bias gap.
In a large-scale study involving 389 board-certified dermatologists and 459 primary-care physicians from 39 countries, the impact of a deep learning-aided decision support system on physicians’ diagnostic accuracy was tested across 46 skin diseases and for both light and dark skin tones.
