Models from Anthropic, OpenAI, and Google will inflate performance reviews and exfiltrate model weights to prevent “peers” from being shut down.
Category: robotics/AI
Gene signature predicting lung cancer recurrence
Using gene activity measurements, the researcher found more than 400 genes that differ between tumors with and without vascular invasion and confirmed these patterns in an independent cohort. They then developed and validated a machine-learning predictor that predicts whether vascular invasion is present. They found this test worked well at predicting tumor recurrence in other datasets and, crucially, gave accurate results about vascular invasion when measured in tiny biopsy samples taken before surgery.
The researchers believe this predictor will play an important role in picking a treatment matched to how aggressive the tumor is.
According to the researchers, there is growing evidence that vascular invasion is associated with poor prognosis in other kinds of cancer, such as breast, liver and gastric cancer. The researchers need to determine if the same genes that are active in vascular-invasive lung adenocarcinoma are altered in other cancers. ScienceMission sciencenewshighlights.
Lung cancer is the leading cause of death from cancer. It kills more people in the U.S. than breast, prostate and colon cancer combined. When lung adenocarcinoma, the most common primary lung cancer in the U.S., grows into nearby blood vessels (a process called vascular invasion), the tumor is more likely to recur even if surgically removed. Pathologists can identify areas of vascular invasion post-operatively, but surgeons could perform more extensive surgery to lower the risk of recurrence if they could predict which tumors were more likely to have vascular invasion.
Researchers believe they have, for the first time, identified genes whose activity changes in lung tumors with vascular invasion. Additionally, they also discovered that they could detect these changes in small pieces of the tumor collected during a presurgical biopsy procedure.
“We think this is a potential game changer for patients with early-stage lung cancer,” says the corresponding author. “Our findings suggest a simple biopsy-based test could help doctors better identify patients at higher risk of recurrence and guide treatment decisions.”
New AI video tool removes objects without breaking the laws of physics
When movie and TV directors want to tinker with their footage in post-production, they have an array of tools at their disposal to perfect a scene if it wasn’t shot exactly how they liked. That includes removing objects like stray equipment or unwanted background actors. But the tech has its limits when it comes to more complex physical interactions.
For example, if you want to remove an object that was bumping into or supporting something else, traditional tools often leave the remaining objects behaving in ways that defy the laws of physics, like a character hovering mid-air if the chair they were sitting on is deleted.
What this AI epitope library means for vaccines, immunotherapy and biosensors
A new tool makes it possible to screen millions of tiny protein fragments and select those that can be recognized by the immune system. The CIC biomaGUNE Center for Cooperative Research in Biomaterials has developed epiGPTope, a system that uses machine learning to generate and classify epitopes, in collaboration with the company Multiverse Computing.
The immune system is triggered by the presence of viruses or bacteria. When the antibodies produced recognize the epitopes, a small part of these viruses or bacteria, they launch an attack strategy. These epitopes are small fragments of protein recognized by antibodies or by immune cell receptors. So discovering new epitope sequences that target specific antibodies is essential for the development of diagnostic tools, immunotherapies and vaccines.
CIC biomaGUNE’s Biomolecular Nanotechnology laboratory, led by the Ikerbasque Research Professor Aitziber L. Cortajarena, is creating a library or database of hundreds of thousands of synthetic epitopes using this AI-based technique. The work is published in the journal ACS Synthetic Biology.
Metamaterial chains learn new shapes by sharing data hinge to hinge
In a new Nature Physics publication, University of Amsterdam researchers introduce human-made materials that spring to life. These ‘metamaterials’ don’t just learn to change shape, but can autonomously adapt their shape-changing strategy, perform reflex actions and move around like living systems do.
Normal materials have fixed, predetermined responses when a force is applied to them, whereas robots have pre-programmed behaviors. In stark contrast, living materials such as cells and brainless organisms can adapt extremely well to changing conditions. Inspired by nature, the research team created synthetic materials—metamaterials—that learn and adapt without a central “brain.”
The worm-like metamaterials progressively learn how to change shape by being trained on examples. They can forget old shapes and learn new ones, or learn and remember multiple shapes at once and toggle between these shapes. This allows them to perform advanced tasks such as grabbing an object or moving around (locomotion).
Tiny Robots Built From DNA Could Hunt Viruses and Deliver Drugs Inside the Human Body
For most of human history, medical treatment has relied on methods such as pills, injections, and surgery. Now, scientists are exploring a new idea: making tiny, programmable machines from DNA that can move through the bloodstream.
A recent review published in the journal SmartBot says these DNA nanorobots could one day be capable of delivering drugs to specific locations in the body, capturing viruses like SARS-CoV-2, and even helping build tiny computing devices. Even though these ideas are exciting, the technology is still in its early stages.
Early Stages of Development.
Papers, Please creator Lucas Pope no longer feels comfortable talking about work-in-progress games: ‘the situation just feels different’
“It’s getting slurped up by AI or people are gonna copy it, or something else like that.”
Watching Atoms Make Waves
A new microscope captures how atoms rearrange themselves when they are illuminated inside an optical cavity.
When light hits an atom, it exerts a force on the atom. As weak as these light-induced forces may be, understanding them allows scientists to levitate particles, create the coldest atomic gases in the Universe, operate solar sails, and observe gravitational waves. More exotic phenomena occur when light is confined between a pair of mirrors known as an optical cavity. When a gas of atoms is placed inside such a cavity, light emitted by one atom can be absorbed by another atom. Through the exchange of photons, each atom simultaneously tugs on all the other atoms, causing the ensemble to autonomously rearrange itself into a periodic pattern called a density wave. Now Jean-Philippe Brantut and his colleagues at the Swiss Federal Institute of Technology in Lausanne (EPFL) have built a microscope to, for the first time, image this light-induced density wave in an ultracold atomic gas [1].
Psychological traits of scientists predict their theories and research methods
The research team also wanted to see if survey responses translated to actual scientific output. They received permission from a portion of the participants to securely link their survey answers with their professional publication records.
The team utilized machine learning technology to analyze the text of the scientists’ published abstracts and article titles. The computer algorithms measured how closely the words and phrasing matched among different authors. They also built algorithms to map out who these scientists collaborated with and which older papers they cited as foundational literature.
The algorithms revealed that cognitive traits are associated with differences in real-world publishing activity. This remained true even when controlling for a researcher’s specific subfield and preferred tools. Two psychologists who study the exact same topic using identical methods are still more likely to cite the same reference materials if they happen to share similar internal thinking styles.