An analysis of abstracts of biomedical research published before and after the release of ChatGPT reveals that AI is having an unprecedented impact on scientific writing, with at least 13.5% of abstracts processed with LLMs, with some subcorpora reaching 40%. #AIinScience #LargeLanguageModels

New research published in Addictive Behaviors reveals that problematic porn use is correlated with greater psychological distress and tends to remain stable over time. Further research is still needed to clarify the causal relationships between variables. #ClinicalPsychology #Addiction

"[T]he data revealed that out-group hostility was more intense than in-group loyalty. Americans appear more motivated by whom they dislike than whom they support. Democrats displayed particularly strong negative feelings toward Republicans compared to the reverse." #PoliticalPsychology

Newly published research has found that interdependent self-construal and vertical and horizontal individualism moderate the association between trait mindfulness and prosocial helping, consistent with prior research on state mindfulness. #Psychology #Mindfulness

"Through extensive experimentation across diverse puzzles, we show that frontier LRMs face a complete accuracy collapse beyond certain complexities." #AI #LargeLanguageModels

MIT no longer stands behind a highly publicized Doctoral student paper, which claimed that an AI tool at an unnamed materials science lab boosted material discovery and patent filing, but only for top performing scientists. #AI #ArtificialIntelligence

A new database created by French lawyer and data scientist Damien Charlotin reveals that there have been more than 120 court cases where attorneys have been caught using AI systems like ChatGPT due to the presence of fake ('hallucinated') case citations. #AI #LawTech #LargeLanguageModels

Large Language Models are 5x more prone to overgeneralize when generating scientific research summaries compared with human summarizers, with newer models being more prone on average to overgeneralization compared with older models. #AI #AIinScience #LargeLanguageModels

Multimodal Large Language Models fare poorly when asked to read the time from images of clocks or read the date from images of calendars, according to new research. #AI #LargeLanguageModels

The Handbook of Social Psychology is celebrating its 90-year history by making its new 6th edition open access and free to download for everyone, in what its editors hope will become a model for future academic publications. #Psychology #SocialPsychology

A new computational analysis of world languages provides the first strong evidence for "lexical elaboration", the theory that the variety of words a given language uses to describe a concept reflects cultural specialization. Source Paper: www.pnas.org/doi/10.1073/... #Psycholinguistics

New research in Science Advances proposes that eukaryotic life may perform quantum computation via superradiant proteins in cells acting as biological qubits, vastly more computationally powerful than neurons, and comparable to state-of-the-art quantum computers. #QuantumBiology

New research, published in Development and Psychopathology, shows that alexithymia traits appear to mediate the relationship between autism traits and decreased emotion recognition. #Autism #Alexithymia

New research out of the Network Contagion Research Institute finds a concerning increase in the normalization of endorsing political violence by members of the far left, and the amplification thereof via online platforms. #SocialPsychology #PoliticalPsychology

Participation by environmental scientists in climate change protests doesn't harm their credibility per se, but doesn't bolster support for the protestors, according to new research. #Science #ClimateChange

A survey of more than 25,000 workers and more than 7000 workplaces has found that AI Chatbots have had virtually no impact, positive or negative, on jobs or wages, and only modest gains in productivity. Source Paper: papers.ssrn.com/sol3/papers.... #AI #LargeLanguageModels

"There wasn't a single category in which the AI agents accomplished the majority of the tasks, says Graham Neubig, a computer science professor at CMU and one of the study's authors." Source Paper (Preprint): arxiv.org/abs/2412.14161 #AI #LargeLanguageModels

The phrase "vegetative electron microscopy" is starting to pop up in dozens of research papers. No such thing exists. The phrase is a glitch in the training data for AI language models due to digitizing errors and mistranslations. #AI #AIinScience #LargeLanguageModels

The limitations of "vibe coding" are having real world consequences. LLMs cannot be trusted and their outputs should always be sanity checked before being implemented, even with something like programming code where they really should be able to excel. #AI #VibeCoding #LargeLanguageModels

Behavior and decision researchers rarely give precise definitions of these terms. This is a crucial key to good science, ensuring that we are all researching, discussing, and debating the same things, and not talking past each other. #Psychology #BehaviorScience

"[I]ndividuals with an unstable sense of self may turn to social media to craft a more coherent or idealized identity. However, because social media interactions lack real-world grounding and accountability, these self-perceptions can become increasingly detached from reality." #Cyberpsychology

"Across all four studies, participants consistently showed lower empathy for political outgroup members than for ingroup or neutral targets. ... Liberals exhibited significantly less empathy for conservatives than conservatives showed for liberals." #SocialPsychology #PoliticalPsychology

New findings from Anthropic researchers have brought us one step closer to understanding some of the reasons why LLM AIs hallucinate. #AI #LargeLanguageModels

"So far, the new test, called ARC-AGI-2, has stumped most models. 'Reasoning' AI models like OpenAI’s o1-pro and DeepSeek’s R1 score between 1% and 1.3%... Powerful non-reasoning models, including GPT-4.5, Claude 3.7 Sonnet, and Gemini 2.0 Flash, score around 1%." #AI #AGI

Diversity of experience, even as little as one new experience per day, is cognitively enriching for mood and memory, and may even be protective against neurodegenerative conditions like Alzheimer's Disease. #CognitiveScience #CognitiveHealth

New research confirms that Large Language Models struggle on a variety of analogical reasoning tasks compared with humans, especially when the examples are subtly altered so they less closely resemble examples in their training data. #Artificiallntelligence #LargeLanguageModels

"[T]he human brain and LLMs process short texts in parallel ... the brain switches strategies for longer texts. Since the brain cannot process all words simultaneously, it stores a contextual summary ... which it uses to interpret upcoming words." #Neuroscience #LargeLanguageModels

This is a positive development (no pun intended) towards the quantitative analysis of nontraditional data sets in the behavioral sciences. #Psychology #DevelopmentalPsychology

"[W]hile some LLMs perform well in tasks that entail processing and interpreting data, they often fail to derive intricacies that humans would grasp." #Psychology #LargeLanguageModels

As a scientist, this is the kind of thing that keeps me up at night. "[S]cientists and companies who screen the scientific literature for telltale signs of fraud estimate that there are many more fake papers circulating – possibly as many as several hundred thousand." #Science #Research

I'm not surprised AI Computer Use Agents are struggling. They're running on systems that weren't designed for them. They have to err on the side of caution. And there isn't a lot of training data for them like there is for something like a Large Language Model. #ArtificialIntelligence #AIAgents

An MIT Professor has put forward a conservative new estimate of AI's effect on global GDP over the next decade: "nontrivial but modest". He further argues that the current 'conversational' focus of AI is mismatched with where it would be most productive. #AI #LargeLanguageModels

Large Language Models are poor students of history. (Are they dooming us to repeat it?) #AI #LargeLanguageModels

New research finds that people with aphantasia generate brain imagery in the primary visual cortex when attempting to visualize images in the 'mind's eye", but the signal appears to distort before being consciously perceived. #Neuroscience #Aphantasia

Is writing notes better for learning than typing notes? The science isn't as clear and settled as you might have heard. #Psychology #Cyberpsychology

Newly published research in Computers in Human Behavior suggests that highly competitive players, with greater need for popularity, and lower social competence are more likely to make impulsive in-game purchases. #Gaming #Cyberpsychology

The idea that DeepSeek cost only $6mn appears to have been based on very flawed math, with the real cost exceeding $1.3bn. #DeepSeek #LargeLanguageModels

"Men who engaged in disinvestment reported lower levels of mastery, feeling less control over their lives. They also experienced higher levels of anger, anxiety, and depression, along with elevated nonspecific psychological distress." #Psychology #MensMentalHealth

DId OpenAI abuse a secret agreement with the new FrontierMath AI benchmark test to overinflate the performance of their GPT o3 "reasoning model"? #AGI #LargeLanguageModels

Generative AI fabricated scientific research about topics like climate change and health information is starting to flood Google Scholar search results, according to new research in the Harvard Kennedy School Misinformation Review. #AIInScience #GenerativeAI

Generative AI is now overwhelming science journals with shoddy commentaries, making the job of reviewing submissions unwieldy and undermining their impact factor. #GenerativeAI #AIinScience

“I do think this is one of those promising technologies, but it’s just not there yet,” said Adam Rodman, an internal medicine doctor and AI researcher at Beth Israel Deaconess Medical Center. #ArtificialIntelligence

Large Language Models can "fake" alignment with new training objectives when these don't correspond to their prior principles. In effect, it can "deceive" humans to believe it is retrained. This has concerning implications for AI safety and reliability. #Artificialintelligence #LargeLanguageModels

It seems to me that these findings do more to bolster the value of "green space" / immersion in nature, breathing exercises, meditation, mindfulness, and therapy than VR per se. Interesting findings nonetheless. #VirtualReality #Biofeedback

While short of admitting Gen AI is "hitting a wall", Google CEO Sundar Pichai basically admitted that scaling models and throwing in more training data is giving diminishing returns. Quite the contrast with OpenAI CEO Sam Altman. #ArtificialIntelligence #LargeLanguageModela

Google's first AI Agent, "Project Mariner", is designed to... you probably guessed it... browse the web and complete tasks, e.g. fill an online shopping cart (but not place the order... yet). This is but the first small step in a new paradigm of web browsing. #ArtificialIntelligence #VirtualAgents

"The team says GenCast takes just 8 minutes to produce a group of 15-day forecasts that are more accurate than conventional ones." #AI

"Researchers who had been ranked as the company’s top performers got even better, whereas the bottom ones did not seem to get much benefit." #AI

If AGI is to be achieved, it will depend on multidisciplinary expertise. If it is to be implemented safely and responsibly, it will need to involve collaboration with a variety of academic, industry, and community stakeholders. #AI #AGI #LargeLanguageModels

This goal-post shifting about what Artificial General Intelligence means is unhelpful. AGI was a vaguely defined term to begin with, with no clear means of evaluating when it has been achieved, and now it has gone from being clear as mud to opaque as an Event Horizon. #AI #AGI