Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: robotics/AI

Clinically informed AI outperforms foundation models in spinal cord disease prediction

Cervical spondylotic myelopathy (CSM) refers to spinal cord compression from arthritis in the neck and is the leading cause of spinal cord dysfunction in older adults. CSM is a chronic, progressive condition that can cause neck pain, muscle weakness, difficulty walking and other debilitating symptoms. While the diagnosis is sometimes clear, often the diagnosis can take years because symptoms aren’t recognized until the later stages, and by then, treatment options are limited.

A multidisciplinary team of surgeon-scientists, computer scientists and researchers at WashU developed an artificial intelligence (AI)-based approach that could help clinicians screen for and diagnose CSM up to 30 months earlier, opening new opportunities for earlier treatment. The findings are published in npj Digital Medicine.

Salim Yakdan, MD, a postdoctoral research fellow in the Taylor Family Department of Neurosurgery at WashU Medicine, and Ben Warner, a doctoral student in computer science and engineering at the McKelvey School of Engineering, co-first authors on the research, used seven different AI models to analyze large datasets containing electronic health record data of more than 2 million people with and without CSM. The models examined patterns of health-care interactions, such as tests and diagnoses, recorded in electronic health records to spot patients whose medical histories resemble those already diagnosed with CSM, helping to flag individuals who may be at higher risk.

Viewing Neural Networks Through a Statistical-Physics Lens

Statistical physics is shedding light on how network architecture and data structure shape the effectiveness of neural-network learning.

Machine-learning technologies have profoundly reshaped many technical fields, with sweeping applications in medical diagnosis, customer service, drug discovery, and beyond. Central to this transformation are neural networks (NNs), models that learn patterns from data by combining many simple computational units, or neurons, linked by weighted connections. Acting collectively, these neurons can process data to learn complex input–output relationships. Despite their practical success, the fundamental mechanisms by which NNs learn remain poorly understood at a theoretical level. Statistical physics offers a promising framework for exploring central questions in machine-learning theory, potentially clarifying how learning depends on the layout of the network—the NN architecture—and on statistics of the data—the data structure (Fig. 1).

Three recent papers in a special Physical Review E collection (See Collection: Statistical Physics Meets Machine Learning — Machine Learning Meets Statistical Physics) provide significant insights into these questions. Francesca Mignacco of City University of New York and Princeton University and Francesco Mori of the University of Oxford in the UK derived analytical results on the optimal fraction of neurons that should be active at a given time [1]. Abdulkadir Canatar and SueYeon Chung of the Flatiron Institute in New York and New York University investigated the influence of the precision with which a network is “trained” on the amount of data the NN can reliably decode [2]. Francesco Cagnetta at the International School for Advanced Studies in Italy and colleagues showed that NNs whose structure mirrors that of the data learn faster [3].

HEART benchmark assesses ability of LLMs and humans to offer emotional support

Large language models (LLMs), artificial intelligence (AI) systems that can process human language and generate texts in response to specific user queries, are now used daily by a growing number of people worldwide. While initially these models were primarily used to quickly source information or produce texts for specific uses, some people have now also started approaching the models with personal issues or concerns.

This has given rise to various debates about the value and limitations of LLMs as tools for providing emotional support. For humans, offering emotional support in dialogue typically entails recognizing what another is feeling and adjusting their tone, words and communication style accordingly.

Researchers at Hippocratic AI, Stanford University, University of California San Diego and University of Texas at Austin recently developed a new structured method to evaluate the ability of both LLMs and humans to offer emotional support during dialogues marked by several back-and-forth exchanges. This framework, dubbed HEART, was introduced in a paper is published on the arXiv preprint server.

How AI can improve the quality of peer review

A new AI coach for scientists has been shown to significantly improve the quality of peer reviews, making them clearer and more helpful for authors. Peer review is essential to ensuring the integrity of scientific publications, but many researchers are dissatisfied with the quality of the feedback they receive. Common complaints include vague, short, and unhelpful reviews. For example, in a survey of 11,800 researchers, only 55.4% of respondents reported being satisfied with the quality of the feedback. The problem is exacerbated by the sheer volume of papers, which has left reviewers feeling overwhelmed.

But help for stressed-out reviewers may be at hand. A team of researchers has developed the Review Feedback Agent, a system that uses five large language models to scan reviews and provide private feedback to reviewers before the authors see them. They trained their AI reviewer by carefully prompting existing large language models, as they explain in a paper published in Nature Machine Intelligence.

The researchers tested their system in the paper review cycle before ICLR 2025, a leading conference in deep learning and machine learning. They randomly assigned around 20,000 reviews to receive AI feedback shortly after they were written. These automated “reviews of the reviews” were then sent back to the human reviewers as private feedback. Another 20,000 were placed in a control group that received no feedback at all.

When light ‘thinks’ like the brain: The connection between photons and artificial memory

An international study has revealed a surprising connection between quantum physics and the theoretical models underlying artificial intelligence. The study results from a collaboration between the Institute of Nanotechnology of the National Research Council (Cnr-Nanotec), the Italian Institute of Technology (IIT), and Sapienza University of Rome, together with international research institutions. The research paper was published recently in the journal Physical Review Letters.

Italian researchers show that identical photons propagating within optical circuits spontaneously behave like a Hopfield Network, one of the best-known mathematical models used to describe the associative memory mechanisms of the human brain.

“Instead of using traditional electronic chips, we exploited quantum interference —the phenomenon that occurs in photonic chips when particles of light overlap and interact with one another to encode and retrieve information,” explains Marco Leonetti, coordinator and corresponding author of the study, senior researcher at Cnr-Nanotec and affiliated with the Center for Life Nano-and Neuro-Science at the Italian Institute of Technology (IIT) in Rome. “In this system, photons are not merely carriers of data, but themselves become the ‘neurons’ of an associative memory.”

AI develops easily understandable solutions for unusual experiments in quantum physics

Researchers at the University of Tuebingen, working with an international team, have developed an artificial intelligence that designs entirely new, sometimes unusual, experiments in quantum physics and presents them in a way that is easily understandable for researchers. This includes experimental setups that humans might never have considered. The new AI doesn’t just create a single design proposal; instead, it writes computer code that generates a whole series of physical experiments, that is, groups of experiments with similar outputs. The study has been published in the journal Nature Machine Intelligence.

The newly developed AI uses a programming language that researchers can easily understand. This allows them to figure out the underlying idea behind the AI’s processes much more easily than before. “AI systems usually deliver their solutions without explaining how they work,” says Mario Krenn, Professor of Machine Learning in Science at the University of Tuebingen and senior author of the study. “We scientists have to try to understand the solutions afterward. This often took us days or weeks—if we understood them at all.”

1Campaign platform helps malicious Google ads evade detection

A newly identified cybercrime service known as 1Campaign is enabling threat actors to run malicious Google Ads that remain online for extended periods while evading scrutiny from security researchers.

1Campaign is a cloaking service that passes Google’s screening process and shows malicious content only to real potential victims. Security researchers and automated scanners are served benign white pages.

The operation has been active for at least three years and is managed by a developer using the name ‘DuppyMeister,’ according to a report from data security company Varonis.

The AI Tsunami is Here & Society Isn’t Ready | Dario Amodei x Nikhil Kamath | People by WTF

I sat down with Dario Amodei in Bangalore. He built Claude, but he started as a biologist looking for a tool to cure disease. Today, he’s at the helm of an AI revolution that he compares to a tsunami society is actively ignoring. We got into the heavy stuff: why Anthropic secretly withheld a working model before ChatGPT existed, whether AI is on the verge of consciousness, and if outsourcing our thinking is going to make humans measurably stupider. Dario makes the case that coding is a dying skill, critical thinking is our last real edge, and the absurd concentration of power in AI right now is a massive problem, even though he’s one of the people holding it.

00:00 Introduction.
06:13 Scaling laws explained simply.
13:27 Trust, humility, and corporate motives.
22:44 Using Claude personally, AI knowing you.
31:03 Rich people criticizing their own system.
37:05 India’s role and IT partnerships.
44:15 Will AI surpass humans at everything.
50:17 Career advice for young Indians.
56:38 Open source vs closed AI models.
1:02:40 Biotech as the next big bet.

#NikhilKamath Co-founder of Zerodha and Gruhas.
Host of ‘WTF is’ & ‘People By WTF’ Podcast.
Twitter: https://twitter.com/nikhilkamathcio/
Instagram: / nikhilkamathcio.
LinkedIn: https://www.linkedin.com/in/nikhilkam… / nikhilkamathcio #Darioamodei LinkedIN– / dario-amodei X — https://twitter.com/DarioAmodei Instagram — / dario.amodei Watch ‘WTF is’ Podcast on Spotify https://tinyurl.com/4nsm4ezn Watch ‘People by WTF’ Podcast on Spotify https://tinyurl.com/yme92c59 Watch ‘WTF Online’ on Spotify https://tinyurl.com/4tjua4th #WTFiswithnikhilkamath #PeopleByWTF #WTFOnline.
Facebook: / nikhilkamathcio.

#Darioamodei.
LinkedIN-/ dario-amodei.
X — https://twitter.com/DarioAmodei.
Instagram — / dario.amodei.

Watch ‘WTF is’ Podcast on Spotify.
https://tinyurl.com/4nsm4ezn.

Watch ‘People by WTF’ Podcast on Spotify.

Continue reading “The AI Tsunami is Here & Society Isn’t Ready | Dario Amodei x Nikhil Kamath | People by WTF” | >

Rapid Evolution of Complex Multi-mutant Proteins

The researchers developed MULTI-evolve, a framework for efficient protein evolution that applies machine learning models trained on datasets of ~200 variants focused specifically on pairs of function-enhancing mutations.

Published in Science, this work represents the first lab-in-the-loop framework for biological design, where computational prediction and experimental design are tightly integrated from the outset, reflecting our broader investment in AI-guided research.

Our insight was to focus on quality over quantity. First identify ~15–20 function-enhancing mutations (using protein language models or experimental screens), then systematically test all pairwise combinations of those beneficial mutations. This generates ~100–200 measurements, and every one is informative for learning beneficial epistatic interactions.

We validated this computationally using 12 existing protein datasets from published studies. Training neural networks on only the single and double mutants, we found models could accurately predict complex multi-mutants (variants with 3–12 mutations) across all 12 diverse protein families. This result held even when we reduced training data to just 10% of what was available.

Training on double mutants works because they reveal epistasis. A double mutant might perform better than the sum of its parts (synergy), worse than expected (antagonism), or exactly as predicted (additivity). These pairwise interaction patterns teach models the rules for how mutations combine, enabling extrapolation to predict which 5-, 6-, or 7-mutation combinations will work synergistically.

We then applied MULTI-evolve to three new proteins: APEX (up to 256-fold improvement over wild-type, 4.8-fold beyond already-optimized APEX2), dCasRx for trans-splicing (up to 9.8-fold improvement), and an anti-CD122 antibody (2.7-fold binding improvement to 1.0 nM, 6.5-fold expression increase). For dCasRx, we started with a deep mutational scan of 11,000 variants, extracted only the function-enhancing mutations, and tested their pairwise combinations—demonstrating the value of strategic data curation for efficient engineering.

Each required experimentally testing only ~100–200 variants in a single round to train models that accurately predicted complex multi-mutants, compressing what traditionally takes 5–10 iterative cycles over many months into weeks. Science Mission sciencenewshighlights.

Continue reading “Rapid Evolution of Complex Multi-mutant Proteins” | >
/* */