Japanese researchers took inspiration from our eyes, creating an artificial retina that could give today’s machine vision systems an upgrade.
The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
Posted in information science, mathematics, robotics/AI | Leave a Comment on The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
Recent generations of frontier language models have introduced Large Reasoning Models (LRMs) that generate detailed thinking processes before providing answers. While these models demonstrate improved performance on reasoning benchmarks, their fundamental capabilities, scal-ing properties, and limitations remain insufficiently understood. Current evaluations primarily fo-cus on established mathematical and coding benchmarks, emphasizing final answer accuracy. How-ever, this evaluation paradigm often suffers from data contamination and does not provide insights into the reasoning traces’ structure and quality. In this work, we systematically investigate these gaps with the help of controllable puzzle environments that allow precise manipulation of composi-tional complexity while maintaining consistent logical structures. This setup enables the analysis of not only final answers but also the internal reasoning traces, offering insights into how LRMs “think”. Through extensive experimentation across diverse puzzles, we show that frontier LRMs face a complete accuracy collapse beyond certain complexities. Moreover, they exhibit a counter-intuitive scaling limit: their reasoning effort increases with problem complexity up to a point, then declines despite having an adequate token budget. By comparing LRMs with their standard LLM counterparts under equivalent inference compute, we identify three performance regimes: low-complexity tasks where standard models surprisingly outperform LRMs, medium-complexity tasks where additional thinking in LRMs demonstrates advantage, and high-complexity tasks where both models experience complete collapse. We found that LRMs have limitations in exact computation: they fail to use explicit algorithms and reason inconsistently across puzzles. We also investigate the reasoning traces in more depth, studying the patterns of explored solutions and analyzing the models’ computational behavior, shedding light on their strengths, limitations, and ultimately raising crucial questions about their true reasoning capabilities.
*Equal contribution. †Work done during an internship at Apple.
Making a discovery with the potential for innovative applications in pharmaceutical development, a West Virginia University microbiology student has found a long sought-after fungus that produces effects similar to the semisynthetic drug LSD, which is used to treat conditions like depression, post-traumatic stress disorder and addiction.
Corinne Hazel, of Delaware, Ohio, an environmental microbiology major and Goldwater Scholar, discovered the new species of fungus growing in morning glory plants and named it Periglandula clandestina.
Hazel made the discovery while working in the lab with Daniel Panaccione, Davis-Michael Professor of Plant and Soil Sciences at the WVU Davis College of Agriculture and Natural Resources. She was studying how morning glories disperse protective chemicals called “ergot alkaloids” through their roots when she saw evidence of a fungus.
Mixed urinary incontinence presents a clinical conundrum. Patients with mixed urinary incontinence report symptoms of both stress incontinence (loss of urine with exertion) and urge incontinence (loss of urine with urgency). Mixed urinary incontinence is a combination of the two that affects 37% of women older than age 65 years.1 The personal and societal costs of incontinence are significant. In women with symptoms of severe urinary incontinence, the cost of supplies, laundry, and dry cleaning range from $900 to $4000 annually.2 By 80 years of age, 20% of women will undergo surgery for stress or mixed urinary incontinence.3 Physical and behavioral therapy improves both incontinence types, and medications are standard treatment for urgency urinary incontinence. When conservative therapies fail, conventional guidance has been to treat the urgency prior to the stress component of mixed incontinence, because anti-incontinence surgical procedures can worsen urgency incontinence, and many urgency treatments are medical rather than surgical.4-7 Another strategy has been to treat whichever symptom is dominant.8
A previously published randomized trial of patients with mixed urinary incontinence compared midurethral sling plus behavioral and physical therapy vs sling alone. Findings from the Effects of Surgical Treatment Enhanced With Exercise for Mixed Urinary Incontinence (ESTEEM) trial revealed that both groups, with or without behavioral and physical therapy, reported improved urgency symptoms, findings that substantiated prior cohort studies.9-11 While the original hypothesis of ESTEEM was that treating both components of mixed urinary incontinence with behavioral and physical therapy plus sling would result in better patient outcomes, ESTEEM revealed that urgency symptoms can improve with the midurethral sling alone, challenging previously held beliefs about the impact of anti-incontinence surgeries worsening the urgency component of mixed incontinence.
In this issue of JAMA, investigators report the results of an important trial that is the next natural step in exploring how best to treat mixed urinary incontinence.12 The Treatment for Mixed Urinary Incontinence: Midurethral Sling vs Botox A (MUSA) is a randomized clinical trial of 137 patients with mixed urinary incontinence and moderate bother from both stress and urge symptoms randomized to either the midurethral sling or 100 U of onabotulinumtoxinA.12 Participants had an average number of 7 leakage episodes a day, representing patients severely affected by incontinence. Importantly, patients previously had unsuccessful conservative interventions, including medications. The investigators hypothesized that treating the urgency component of mixed urinary incontinence with onabotulinumtoxinA would result in better outcomes than focusing on the stress component with a midurethral sling.
A disposable, electronic “tattoo” that measures mental workload could be used to enhance safety in high-pressure jobs, researchers say.
Australian researchers are turning to nature for the next computing revolution, harnessing living cells and biological systems as potential replacements for traditional silicon chips. A new paper from Macquarie University scientists outlines how engineered biological systems could solve limitations in traditional computing, as international competition accelerates the development of “semisynbio” technologies.
Living computers, organs-on-a-chip, data storage in DNA and biosecurity networks that detect threats before they spread—these aren’t science fiction concepts but emerging realities. A team from Macquarie University and the ARC Center of Excellence in Synthetic Biology (COESB) has explored this convergence of biological and digital technologies in a Perspective paper published in Nature Communications.
The Macquarie University authors—Professor Isak Pretorius, Professor Ian Paulsen and Dr. Thom Dixon (who are also affiliated with the ARC Center of Excellence in Synthetic Biology), Professor Daniel Johnson and Professor Michael Boers—draw on decades of combined experience to explain why harnessing bio-innovation can proactively shape the future of computing technology.
Using advanced computational modeling, a research team led by the University of Oxford, working in partnership with the Instituto Superior Técnico at the University of Lisbon, has achieved the first-ever real-time, three-dimensional simulations of how intense laser beams alter the “quantum vacuum”—a state once assumed to be empty, but which quantum physics predicts is full of virtual electron-positron pairs.
Excitingly, these simulations recreate a bizarre phenomenon predicted by quantum physics, known as “vacuum four-wave mixing.” This states that the combined electromagnetic field of three focused laser pulses can polarize the virtual electron-positron pairs of a vacuum, causing photons to bounce off each other like billiard balls—generating a fourth laser beam in a “light from darkness” process. These events could act as a probe of new physics at extremely high intensities.
“This is not just an academic curiosity—it is a major step toward experimental confirmation of quantum effects that until now have been mostly theoretical,” said study co-author Professor Peter Norreys, Department of Physics, University of Oxford.
A new study reveals that the brain’s default mode network is made up of distinct anatomical types that support both internal thoughts and external processing. This structural diversity helps explain the network’s role in everything from memory to imagination.
The discovery could reshape how we study psychedelic compounds in nature and medicine.