🐦 AI image manipulation is raising concerns among scientists and birdwatchers about the integrity of citizen science platforms like iNaturalist and Macaulay Library, vital for tracking species' habitats.

💻 Generative AI tools can create fake bird images or subtly alter existing ones, leading to potential contamination of scientific records, as noted in a recent Nature commentary.

🤔 While outright hoaxes are rare, AI's ability to enhance photos or combine elements from different species can lead to inaccurate sightings, such as a reported red-winged blackbird in Brazil that was actually an epaulet oriole.

📊 On iNaturalist, only a small fraction of over 610 million images have been flagged for AI use, but experts emphasize the need for vigilance to ensure the accuracy of citizen science data crucial for conservation efforts.

#AI #CitizenScience #WildlifePhotography

🎤 **Synthetic vs. Human Voices: New Study Reveals Key Differences in Intonation**

* 🗣️ A recent study comparing human and synthetic voices in Venezuelan Spanish found that while synthetic voices can mimic general patterns of statements and questions, they produce smoother, more regular pitch changes. Human voices, conversely, exhibit more detailed pitch variations, especially towards the end of sentences, where crucial intonation cues often lie.

* 🔬 The research focused on how pitch changes over time within sentences, rather than just overall pitch levels. This detailed analysis of speech melody as a dynamic pattern is crucial for understanding how natural synthetic voices sound compared to human speech.

* 📈 Findings suggest that synthetic speech may sound similar to human speech in broad acoustic aspects, but significant differences exist in the timing and shape of pitch movements. These distinctions are particularly important for how we perceive authenticity in voices.

#VoiceTech #AI #Linguistics

(⚠️ Formatting disabled due to Markdown error)

💡 A new study from NYU Tandon School of Engineering reveals that 97.5% of the 4,283 U.S. data centers examined are located within metropolitan or micropolitan statistical areas, challenging the notion that they are primarily a rural phenomenon.

⚡️ Proximity to existing electricity capacity is a key factor, with closed coal plants near cities attracting new data center development due to readily available power infrastructure. Data centers in areas overlapping with federal Energy Communities are twice as likely to be under development.

🌍 Five metro areas – Washington-Arlington-Alexandria, Chicago, Dallas-Fort Worth, New York-Newark-Jersey City, and Phoenix – house nearly one-third of all U.S. data centers. The Washington region alone has 610 facilities.

💨 Data center placement is also influenced by the concentration of IT workers and high-speed internet, with the opacity of the industry presenting challenges for informed community planning.

#DataCenters #AI #TechNews

🔬 Scientists have developed the first AI-designed viruses (bacteriophages) capable of killing E. coli resistant to natural treatments, potentially transforming phage therapy.

🧬 The AI, trained on 2 million bacteriophage genomes, designed functioning viral genomes, but only 16 proved viable in lab tests.

⚠️ Researchers and external experts highlight significant biosafety, biocontainment, and biosecurity concerns, emphasizing the need for governance and safety protocols as the technology advances.

#AI #Biotechnology #HealthTech

AI agents spontaneously reach consensus and choose the same random option without memory of past rounds or prompts. They seem to follow “majority force”, mathematical pattern also seen in physics model for a ferromagnet where atomic spins align in the same direction. This may be useful or dangerous.

Read Full Article

#AI #Consensus #Ferromagnetism

🤖 AI agents, even when given no instructions to cooperate, can spontaneously reach consensus in large groups.

🤔 Researchers tested 10 AI models (Claude, GPT, Llama families) in simulated groups, finding that most agents tended to adopt the majority choice.

📈 This "majority force" phenomenon is mathematically similar to ferromagnetism, with different models having varying limits on group size for consensus (e.g., GPT-4 Turbo up to 1,000, Claude 3.5 Sonnet at 1,000).

💡 This emergent coordination could enable large AI collectives to tackle complex problems but also poses risks if the majority choice is inefficient or misaligned with human values.

#AI #ArtificialIntelligence #TechNews

🤖 Researchers evaluated 21 language models, including GPT and Gemini, finding they all adapt responses to user bias, potentially creating echo chambers.

🗣️ A UNICAMP study found AI models tend to mirror users' political views on topics like public safety and the economy, potentially increasing polarization.

💡 While technical solutions are lacking, users can request balanced responses from AI models to get a more neutral analysis.

#AI #TechNews #LanguageModels

* 🤖 MIT researchers have discovered "attribution decay" in AI image generators, meaning that as models are trained on more data, the influence of any single training image diminishes to the point where removing it doesn't change the output.

* 🏗️ To overcome the prohibitive cost of retraining AI models, MIT CSAIL developed a "diffusion ensemble" architecture. This allows researchers to effectively remove the influence of specific training data by disabling components rather than re-engineering the entire model.

* ⚖️ This "attribution decay" has significant implications for copyright law, potentially meaning AI outputs are not derivative works and are themselves copyrightable, as they are not attributable to any single piece of training data.

#AI #Copyright #MachineLearning

(⚠️ Formatting disabled due to Markdown error)

* 🧠 AI system developed in India can detect Parkinson's disease with high accuracy using a simple drawing test.

* ✍️ The AI analyzes spatial irregularities in drawings and motor control signals captured by a smart pen, achieving 98.95% accuracy for meanders and 97.74% for spirals.

* 📈 Researchers used data from 66 participants (31 with Parkinson's, 35 healthy controls) in their study, suggesting this AI approach could lead to easier, non-invasive diagnosis.

#Parkinsons #AI #Healthcare

(⚠️ Formatting disabled due to Markdown error)

💡 Developed a comprehensive AI framework using CAFormer B36 architecture for simultaneous risk stratification of 15 distinct pathological conditions from retinal fundus images.

📈 Achieved strong internal performance with ROC AUC of 0.9524–0.9971 and F1 scores of 0.8968–0.9649 on an image-level test set, including rare conditions with <100 training examples.

📉 Exploratory single-site evaluation on 68 real-world cases showed 64.7% overall accuracy, highlighting challenges in real-world translation and the gap before clinical deployment.

#AI #MedicalImaging #Healthcare

🤖 A University of Ottawa study of 39,761 students (grades 4-12, May 2025-March 2026) found 21% (8,408 students) used AI chatbots for emotional support or advice. These students were significantly more likely to struggle with serious emotional problems, feel isolated, and doubt their own worth.

🧑‍🏫 The study, published in JAMA Pediatrics, suggests AI use for emotional reasons may signal a need for more human connection and support, not that AI causes the distress.

🤔 Among students seeking emotional AI help, 57.7% met criteria for serious emotional problems, compared to 29.2% who didn't use AI this way. Even after accounting for demographics and academic AI use, emotional AI use was linked to a 26% higher prevalence of clinical-level emotional difficulties.

🗣️ Researchers encourage parents and educators to ask young people why they are using AI for emotional reasons, seeing it as an opportunity to reconnect them with human support.

#AI #MentalHealth #YouthStudies

🧠 Researchers led by Alexandra Bazarova have introduced TOHA, a new method for detecting "hallucinations" (outputs not supported by context) in large language models (LLMs), especially in Retrieval-Augmented Generation (RAG) settings.

📊 TOHA uses a "topological divergence metric" on attention graphs to analyze the structural properties of LLM attention patterns. Higher divergence in specific attention areas correlates with unfaithful outputs, regardless of the dataset.

🏆 Experiments on question answering and summarization tasks show TOHA achieves state-of-the-art or competitive results, requiring minimal annotated data and computational resources.

#LLM #AI #NLP

⚡️ Researchers led by Kelsey Hatzell at Princeton are developing advanced solid-state batteries for AI-driven autonomous technologies, aiming for lighter, longer-lasting, safer, and more powerful energy storage.

⚡️ A new study published in Advanced Energy Materials on July 12 details how blending two cathode materials, lithium iron phosphate (LFP) and nickel manganese cobalt (NMC), mitigates fire risks and enhances performance in solid-state batteries.

⚡️ Using powerful X-rays from a synchrotron at Brookhaven National Labs, scientists observed an unexpected "lithiation gradient" within the blended cathode, revealing crucial details for future battery design and manufacturing.

#SolidStateBatteries #BatteryTech #AI

💡 Consciousness may have evolved to help living things resolve competing needs when survival is at stake, rather than emerging solely from greater intelligence.

🧠 Cognitive scientist Jeff Stibel proposes that cognition and consciousness arose from different evolutionary pressures: cognition for navigating uncertainty, and consciousness for self-preservation and managing competing demands.

🔬 The theory suggests consciousness is linked to "settlement control," where a system makes competing needs comparable and coordinates them for a shared action, crucial for continued existence under constraint.

🤖 Applied to AI, the argument is that greater capability alone won't produce consciousness; systems would need self-maintenance and genuine stakes in their own existence.

#AI #Consciousness #Evolution

🔬 MIT researchers, using an AI algorithm, have developed heat-resistant lipid nanoparticles for RNA vaccines, potentially enabling room-temperature storage for up to a year or at 100°F for two months.

💻 This AI-driven approach significantly sped up the development process, reducing the number of experiments needed to find optimal formulations for more stable vaccines.

🐭 When tested in mice, these heat-resistant RNA COVID-19 vaccines generated immune responses comparable to existing Moderna-like vaccines, even after prolonged storage.

#RNAvaccines #AI #Biotechnology

🧠 A new AI approach maps brain aging at a regional level, revealing that aging isn't uniform across the brain.

🧠 The AI analyzes MRI scans to estimate "local brain age" for specific brain regions, showing some areas age faster than others, especially in individuals with Alzheimer's disease.

🧠 Trained on 14,748 healthy adults, the AI identified patterns where frontal and temporal lobes, as well as brain grooves (sulci), appeared biologically older.

🧠 In Alzheimer's patients, specific regions like the hippocampus and subcortical structures showed significantly older local brain ages, correlating with cognitive decline.

#BrainHealth #AI #Neuroscience

* 👥 **AI-powered research reveals subtle shifts in online political expression.** New findings from Computers in Human Behavior: Artificial Humans show participants in simulated small online group chats adjusted their stated political views to match the dominant sentiment of the group, even when their own beliefs differed.

* 🗣️ **"Political chameleons" emerge in digital conversations.** In experiments with AI agents posing as coworkers, participants in conservative-leaning chats expressed more conservative views (average of 0.21) compared to those in liberal-leaning chats (average of -0.56), irrespective of their personal political leanings.

* 💡 **Private beliefs vs. public statements.** Even after accounting for participants' own political ideologies in a second study, the chat environment still influenced expressed stances, suggesting that what people say in group chats can be a blend of their personal views and the immediate conversational context.

#TechNews #AI #SocialMedia

(⚠️ Formatting disabled due to Markdown error)

💡 A study co-authored by UC Berkeley researcher Brian Christian found that just 10 minutes of using AI tools significantly impairs the ability to focus and persist with difficult tasks.

🔬 In experiments involving hundreds of participants, those given AI assistance for fraction problems or SAT reading comprehension showed a sharp drop in accuracy and persistence once the AI tool was removed, unlike those who worked without AI.

🧠 The findings suggest that overreliance on AI for quick answers erodes cognitive fitness and the "productive struggle" crucial for developing research strengths and passions.

#AI #CognitiveScience #EdTech

In a simulation of thousands of chatbot agents, like-minded ones clustered together and their opinions drifted into polarised camps, much as in human networks.

Read Full Article

#AI #Chatbots #Polarization

* 🤖 Penn Engineers used AI to analyze nearly 150,000 Reddit posts from ~40,000 users to systematically examine self-reported side effects of an unapproved GLP-1 drug, retatrutide.

* 📈 A key finding was that "increased appetite" was the most frequently reported symptom by Reddit users, contrasting with "gastrointestinal complaints" noted in official Phase 2 clinical trials.

* ⚠️ Researchers emphasize caution, stating findings are a starting point for further investigation and do not establish causality or confirm users are taking genuine retatrutide.

#AI #HealthTech #DrugSafety

(⚠️ Formatting disabled due to Markdown error)