Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: robotics/AI – Page 3

Bioengineers condense protein engineering and testing to a single day

Proteins are critical to life—and to industry. There are countless proteins that could be engineered to treat and even cure serious diseases and cellular dysfunctions. Industrial applications are similarly promising, with proteins increasingly used as enzymes in food manufacturing and in consumer detergents.

While AI can help suggest improvements, each novel protein must still be created in the real world and tested for performance. It is a labor-intensive process that involves constructing the DNA instructions for each protein in yeast or bacteria and growing individual clones for protein production and testing. This can take many days for a single protein of interest and even longer if the protein needs to be tested in mammalian cells, a process that requires retrieving DNA from microbes for transfer to the mammalian cells.

In a new paper, Michael Z. Lin, a professor of neurobiology and of bioengineering in the schools of Engineering and Medicine, and graduate students, Yan Wu in bioengineering and Pengli Wang in chemical engineering, say they have condensed the time-intensive protein building and testing process to just 24 hours.

Reconfigurable Ge-Si photodetector achieves ultrahigh-speed data transmission using low-loss packaging

The rapid growth of large language models is placing increasing demands on data centers, where large volumes of data must be transferred efficiently between servers. Optical interconnects are essential for enabling this communication, but as data rates continue to rise, these systems must deliver higher bandwidth while maintaining low latency and energy efficiency. However, integrating electronic and photonic components remains challenging, as conventional approaches often introduce signal loss, limit interconnect density, and restrict scalability.

As reported in Advanced Photonics Nexus, Dr. Wei Chu and colleagues have developed a reconfigurable germanium–silicon photodetector using a low-loss integration strategy based on fan-out wafer-level packaging (FOWLP). This approach enables seamless integration of electronic integrated circuits and photonic integrated circuits on a single platform without the need for traditional wire bonding, reducing parasitic loss and improving signal integrity.

The system uses a dense network of fine metal interconnects, known as a redistribution layer (RDL), to connect components with high precision. This structure supports high interconnect density—exceeding 102 connections per square millimeter—while maintaining a low insertion loss of less than 0.3 dB/mm at 100 GHz. In addition, the use of benzocyclobutene as a low-dielectric insulating material reduces transmission loss and improves thermal stability for reliable high-frequency operation.

AI shapes the design of the electron-ion collider

Artificial intelligence and machine learning are shaping major design and research decisions for the planned Electron-Ion Collider (EIC), a next-generation nuclear physics research facility that will collide electrons with protons or nuclei to probe matter’s structure.

The EIC—being built at the U.S. Department of Energy’s (DOE) Brookhaven National Laboratory in partnership with DOE’s Thomas Jefferson National Accelerator Facility (Jefferson Lab)—will reveal the inner structure of matter in unprecedented detail. It is the world’s first collider designed with AI and machine learning integrated into both its accelerator and detector systems.

“EIC is a new facility that can take advantage of AI and machine learning from the start,” said Tanja Horn, a professor of physics at The Catholic University of America, and co-chair of AI4EIC, a working group devoted to developing AI for the EIC. “A wide array of AI tools is now available—perfectly timed for the EIC.”

Generalization Dynamics of LM Pre-training

An AI has a limited amount of “capacity” (brainpower). Early in training, it develops quick, shallow circuits to memorize data because that’s the easiest way to get the right answer. Later, it develops complex circuits for actual reasoning. Because space is limited, these two internal systems are constantly competing for control. Whichever type of data the AI happens to be reading in a specific moment determines which circuit wins the battle.

People typically assume that LMs stably mature from pattern-matching parrots to generalizable intelligence during pre-training. We build a toy eval suite and show this mental model is wrong: throughout pre-training, LMs frequently and suddenly hop between parrot-like and intelligence-like modes, i.e. distinct algorithms implemented by distinct circuits. We call this mode-hopping. Across our suite, LMs can suddenly latch onto memorized or in-context patterns instead of in-context learning, use System 1 instead of System 2 thinking, pick up what sounds true instead of what is true, fail at multi-hop persona QA, out-of-context reasoning, and emergent misalignment — then just as suddenly revert and generalize. Mode-hopping is not explained by standard optimization dynamics: it is locally stable and can not be fixed by checkpoint averaging. We instead think of it as a capacity allocation problem: in a capacity-bounded model, generalizable circuits must compete with the shallow ones learned early in training, and the data in each pre-training window decides which circuits win. Our suite provides a cheap set of pre-training monitors and a new lens on generalization. Building upon our insights, we demonstrate three applications: (i) select intermediate pre-training checkpoints that strongly generalize reasoning and alignment, better than the final pre-or mid-training checkpoints, (ii) select pre-training data that controls and stabilizes generalization dynamics, and (iii) test prior generalization predictors, falsifying the monolithic belief that “simpler solutions generalize better”

Building general AI without generalization is doable but meh. We want an intelligence that learns deep, transferable structure, not a parrot that matches shallow patterns. Real generalization would unblock many today’s key open problems: data-efficient (online) learning, shortcut learning, transfer capabilities from verifiable domains (math, coding) to broader non-verifiable yet economically valuable domains, and maintain a coherent character that truly aligns with human values.

The distinction between parrots and intelligence is computational. Parrots repeat in-context patterns; intelligence infers in-context functions. Parrots encode a persona as bags of disconnected facts and traits; intelligence learns a shared persona representation that connects all. Parrots memorize reasoning steps; intelligence forms general reasoning circuits for entity tracking, backtracking, or even for highly abstract concepts like truth.

AI-generated fake citations are flooding scientific literature across publications, scientists warn

The citations at the end of a research paper should represent a solid foundation of existing knowledge about a particular field, a pool of peer-reviewed sources built over years of research and study. However, with the increasing use of AI and large language models in writing research papers, there’s a growing chance that the citation someone clicks on may not even exist, and that the study, the source, or even the researchers themselves could be entirely fake.

In a recent study posted to the arXiv preprint server, researchers audited millions of papers and found that an estimated 146,900 hallucinated citations were present in research papers hosted on four major scientific repositories— arXiv, bioRxiv, SSRN, and PubMed Central. These numbers were for 2025 alone.

The hallucinated citations were not limited to a handful of bad apples but appeared across many papers, each containing a small number of fake references, pointing to a broader pattern of researchers using AI yet failing to fact-check the output.

Stanford CS231N Deep Learning for Computer Vision I 2025

Computer Vision has become ubiquitous in our society, with applications in search, image understanding, apps, mapping, medicine, drones, and self-driving cars. Core to many of these applications are visual recognition tasks such as image classification, localization and detection. Recent developments in neural network (aka “deep learning”) approaches have greatly advanced the performance of these state-of-the-art visual recognition systems. This course is a deep dive into deep learning methods with a focus on end-to-end models for core vision tasks, alongside modern approaches such as transformers, diffusion models, and visual-language models that power today’s AI systems. During the 10-week course, students will learn to implement and train their own neural networks and gain a detailed understanding of cutting-edge research in computer vision. Additionally, the final assignment will give them the opportunity to train and apply multi-million parameter networks on real-world vision problems of their choice. Through multiple hands-on assignments and the final course project, students will acquire the toolset for setting up deep learning tasks and practical engineering tricks for training and fine-tuning deep neural networks. https://online.stanford.edu/courses/cs231n-deep-learning-computer-vision

/* */