Vision-language models (VLMs) embed aligned image-text pairs into a joint space but often rely on deterministic embeddings, assuming a one-to-one correspondence between images and texts. This oversimplifies real-world relationships, which are inherently many-to-many, with multiple captions describing a single image and vice versa. We introduce Probabilistic Language-Image Pre-training (ProLIP), the first probabilistic VLM pre-trained on a billion-scale image-text dataset using only probabilistic objectives, achieving a strong zero-shot capability (e.g., 74.6% ImageNet zero-shot accuracy with ViT-B/16). ProLIP efficiently estimates uncertainty by an "uncertainty token" without extra parameters. We also introduce a novel inclusion loss that enforces distributional inclusion relationships between image-text pairs and between original and masked inputs. Experiments demonstrate that, by leveraging uncertainty estimates, ProLIP benefits downstream tasks and aligns with intuitive notions of uncertainty, e.g., shorter texts being more uncertain and more general inputs including specific ones. Utilizing text uncertainties, we further improve ImageNet accuracy from 74.6% to 75.8% (under a few-shot setting), supporting the practical advantages of our probabilistic approach. The code is available at https://github.com/naver-ai/prolip

Windows-based portable gaming PCs are all the rage at Computex, and MSI came loaded for bear with an update to its Claw design. The Claw 8 AI will launch with a Lunar Lake processor and a larger screen later this year. Adam cornered MSI's System Product Management Director Cliff Chun on the show floor to talk about it. In addition to a new BIOS and drivers created in cooperation with Intel that give the Meteor Lake version of the handheld a claimed 70 percent boost in performance, there will be a Lunar Lake-equipped variant of the 7-inch Claw released. But the big news is the 8-inch Claw 8 AI, which will debut along with Lunar Lake.

The protagonist of the Netflix megahit Squid Game is Seong Gi-hun, an indebted gambler and absentee father who screams, sweats, and strains his way through the very intense experience of watching hundreds of people get straight-up killed--while trying to avoid being killed, and retain some sense of ethics and loyalty, to boot. But we wondered: what was it like to voice Gi-hun in English for the many people who watched Squid Game with the dubbing option turned on? So we asked the voice actor Greg Chun, a veteran of video games and anime who spoke to Slate from his studio in Los Angeles. Our conversation--on the hardest Squid Game scene to dub, the controversy around the Korean-to-English translation, and his time working on Call of Duty--has been edited and condensed for clarity. Rebecca Onion: What was your reaction when you first saw the Squid Game script?

"We are visual beings and knowing how the brain rapidly computes where to look is fundamentally important," said Yale's Marvin Chun, Richard M. Colgate Professor of Psychology, professor of neuroscience and co-author of new research published Dec. 4 in the journal Nature Communications. Eye movements have been extensively studied, and researchers can tell with some certainty where a gaze will be directed at different elements in the environment. What hasn't been understood is how the brain orchestrates this ability, which is so fundamental to survival. In a previous example of "mind reading," Chun's group successfully reconstructed facial images viewed while people were scanned in an MRI machine, based on their brain imaging data alone. In the new paper, Chun and lead author Thomas P. O'Connell took a similar approach and showed that by analyzing the brain responses to complex, natural scenes, they could predict where people would direct their attention and gaze.

Unlike most cells in the rest of our body, the DNA in each of our brain cells is not the same: it varies from cell to cell, caused by somatic changes. This could explain many mysteries--from the cause of Alzheimer's disease and autism to how our personality develops. But much remains unknown, including when these changes arise, their size and locations and whether they are random or regulated. DNA technologies used to study these "copy number variations" (CNVs) in single brain cells have been limited to longer DNA sequences--those above one million base pairs. Now, scientists at Sanford Burnham Prebys Medical Discovery Institute have developed new single-cell analysis approaches wedded to machine learning, allowing the detection of CNVs smaller than one million base pairs.

In the hit movie Avengers: Age of Ultron, the Iron Man shows the'brains' of a computer system to his colleague, the Incredible Hulk. "I mean, look at this! They're like neurons firing," the Hulk exclaims, pointing to a pulsating, blue orb which represented super baddie Ultron's consciousness. We'd like to think that's what neural networks look like too. They are a rising field of artificial intelligence, and a new trend that is coming to a government near you. Neural networks describe a computing technique that closely imitates human brain functions. "By using neural networks, we try to mimic nature's ability to learn how certain things work," Associate Professor Andy Chun from City University of Hong Kong's Department of Computer Science tells GovInsider.

JUST after midnight, the last subway car slips into its sidings in Hong Kong and an army of engineers goes to work. In a typical week, 10,000 people carry out 2600 engineering works across the system – from grinding rough rails smooth and replacing tracks to checking for damage. People might do the work, but they don't choose what needs doing. Instead, each task is scheduled and managed by artificial intelligence. Hong Kong has one of the world's best subway systems.

Today, telepresence robots are priced for purchase or rent anywhere between 5,000 to 200,000, Chun says. While his research suggests the telepresence market is valued at 4 billion, he estimates that it's actually about half when factoring in conservative assumptions. Still, there is a lot of promise in the near future. Chun recorded 20 telepresence robot patents in 2014, and projected 36 patents for 2015. Additionally, the United States remains the dominant market, but Chun is seeing expansion in countries in Asia and the Middle East.