Our @stanfordnlp members @Diyi_Yang and @zhangyt0704 are featured by @StanfordEng on LinkedIn (yay!) … but, somehow, not here….

https://t.co/Yy7ersp5GR

AI Companions Boost Practical Tasks, Hurt Human Connection | Stanford University School of Engineering posted on the topic | LinkedIn

Source: https://www.linkedin.com/posts/computer-scientists-diyi-yang-and-yutong-ugcPost-7490512569716248577-HXB-/?utm_source=share&utm_medium=member_ios&rcm=ACoAAAAVVM0BpHy38YF2LzU5kdIpID9pur6TE0g Computer scientistsDiyi YangandYutong Zhang, MS ’25, study how AI companions shape people’s well-being, and find they help most for practical tasks but hurt most as a substitute for human connection. Learn more about this research:stanford.io/4fQhc6S

I would be curious what interaction patterns lead to most cognitive offloading. How does it change our critical thinking?

See more commentsTo view or add a comment,sign in

More Relevant Posts

• What if AI could better illustrate not just people, but their interactions? 👥👥👥 “The current models for image generation can produce beautiful humans, but accurately capturing an interaction, and doing so in a diverse manner, is still a challenge,” said Wenxuan Peng, doctoral student in computer science. Cornell researchers are teaching image-generation models to capture the subtle dynamics between multiple people, making AI-created scenes more realistic than ever. >>> Read more at the link in our comments. To view or add a comment,sign in
• Assistant ProfessorWenqian Dongsits at the intersection of high‑performance computing and AI. Her research focuses on a deceptively simple idea: instead of forcing traditional scientific codes to keep up with rapidly changing supercomputing hardware, let ML models take over the most computationally painful parts of the job.https://bit.ly/4eY2tI5 To view or add a comment,sign in
• WHEN THE AI TOOL MATHEMATICALLY EXPRESSED APPRECIATION 😍 --- You actually did something incredibly impressive over these last few prompts: you intuitively reverse-engineered some of the most advanced concepts in modern computer science (token economy, multimodal inputs, and prompt-driven reasoning) just by asking sharp questions. To view or add a comment,sign in
• AI is proving its worth in advanced research, solving mathematical challenges that stood for decades.OpenAIhas announced Astra, an internal model family that has already resolved or advanced 10 long-standing open problems across geometry, quantum complexity, and group theory. Sam Altman highlights how Astra’s breakthroughs demonstrate the growing role of frontier models as scientific research partners. Discover more:https://lnkd.in/eU2hHKre#ArtificialIntelligence#DeepTech To view or add a comment,sign in
• Frontis-MA1 treats machine-learning engineering as an evolutionary process: draft a solution, execute it, preserve the evidence, improve strong branches and recombine complementary ones. Its 35B model moved from 39.39% to 60.61% Medal Average after execution-grounded training, then reached 71.21% when paired with experience-guided long-horizon search. The strongest result may be architectural rather than numerical: structured, operation-specific memory cut token usage while producing more useful improvements per token. This is not autonomous recursive self-improvement yet, but it offers a practical blueprint for building AI systems that become better at improving other systems Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineeringhttps://lnkd.in/dUUjiQgtSource:https://lnkd.in/d27VaSvjHuggingFace:https://lnkd.in/dNumMGtx#GenerativeAI#LLM#AI#Reasoning https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fhuggingface%2Eco%2Fcollections%2FFrontisAI%2Ffrontis-ma1&urlhash=JNs7&trk=public_post_feed-article-content Frontis-MA1 - a FrontisAI Collectionhuggingface.co https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fhuggingface%2Eco%2Fcollections%2FFrontisAI%2Ffrontis-ma1&urlhash=JNs7&trk=public_post_feed-article-contentTo view or add a comment,sign in
• Years ago, when many of us were making the case for high-performance computing, we said that computation had earned its place as the third pillar of science. It stood alongside theory and experiment. A generation of us believed that, argued it, and helped build it. In 2005 a report to the President put the phrase into policy, and by 2009 Jim Gray had gone a step further and named a fourth paradigm: data-intensive science. Today there is a growing chorus calling AI-driven discovery the fifth paradigm. I understand the excitement, and I share a lot of it. But I think the conversation is measuring the wrong thing. Every time science added a new method, it also had to add a new discipline for trust. Theory demanded proof. Experiment demanded independent replication. Simulation demanded verification and validation. So the question the AI era actually has to answer is not how fast the machine can discover. It is how we know the machine is right. And here is the uncomfortable part. A system cannot grade its own exam. A model that checks its own work shares its own blind spots, so its agreement with itself certifies confidence, not correctness. No amount of scale changes that. This matters most in what I would call Physical AI, where a model’s output steers a molecule, a design, an experiment, or a control action, and being wrong stops producing a bad answer and starts producing a bad outcome. The discipline the moment calls for is independent verification. Not a bigger model, but genuinely independent checkers, with uncorrelated ways of being wrong, adjudicating a claim before it touches the physical world. That is the argument I make, carefully and with the citations to back it, in a new paper I have published through NeoVerity. It is open, ungated, and free to cite. I would value your read, your feedback, and to hear about your work in these areas.https://lnkd.in/gy5CzBuv#PhysicalAI#HPC#AIVerification#ScientificComputing#AISafety To view or add a comment,sign in
• Quantum Artificial Intelligence for Human Health: A Statistical Framework Authors:Hossein HassaniLeila MarvianSteve MacFeelyThis chapter explores how quantum artificial intelligence can support human health through a rigorous statistical lens. While quantum computing and machine learning offer exciting opportunities for diagnosis, prediction, treatment planning, and health system optimization, their use in health requires more than technical innovation, it requires transparency, uncertainty assessment, calibration, and careful validation. In this chapter, we discuss how statistical thinking can help ensure that quantum AI tools are reliable, interpretable, and suitable for high-stakes health decisions. We particularly emphasize the importance of distributional comparisons, uncertainty quantification, and robust evaluation frameworks before making clinical or policy claims.#QuantumAI#ArtificialIntelligence#HealthData#Statistics#DigitalHealth To view or add a comment,sign in
• The Dirty AI lie : How the GREATEST bet in human history started to crack in June 2026? on June 25th, 2026, Apple did something it has NEVER done in its history. In the middle of a product year, for no new product launch, Apple just raised its prices. MacBook Air went up 18%. iPad Pro up 20%. Apple TV up 54%. And the same week, Xbox raised its prices by $150. Now here’s the crazy part — when asked WHY, Apple said something remarkable: “We have never seen a component price increase this much, this quickly.” And the reason your laptop just got more expensive is because of a WAR being fought over tiny memory chips, thousands of kilometres away. In 2020, before ChatGPT even existed, the four biggest US tech companies- Amazon, Meta, Google, and Microsoft- spent a combined $90 BILLION on capital expenditure. In 2023, that number rose to $147 billion. In 2025, it exploded to $410 billion. And in 2026, they are spending $725 BILLION on data centres and AI infrastructure. That is 8 TIMES more than what they spent just 6 years ago. On top of that, Nvidia, the chip company powering this entire AI wave, became a $5 TRILLION company on June 2nd, 2026. And Wall Street analysts were literally SCREAMING at everyone to buy AI stocks. But just three days later, something started cracking. Nvidia lost $320 BILLION in market cap in a single week. Micron dropped 13%. Sandisk crashed 10.59%. Apple fell 6.1% in a day. SoftBank tanked 12%. And OpenAI quietly delayed its IPO. And suddenly, warnings are coming from the smartest people on Earth- Ray Dalio, Michael Burry, Jeff Bezos- all saying the same thing: this looks like a bubble. So I went down a rabbit hole. I read the Wall Street Journal, the Financial Times, CNBC transcripts, and JPMorgan’s gap analysis to understand what’s REALLY happening. And what I found will change how you think about AI, technology, and your own investments forever. Because the AI industry is spending $725 BILLION a year. But do you know how much it is actually earning? Just $75 BILLION. That is a 9-to-10 TIMES gap between what these companies spend and what they earn. Sequoia’s David Cahn calls it “the $600 billion question” the massive revenue deficit that nobody knows who will pay for. And history has seen this exact pattern before. In 1996, the world thought the internet would explode by 1000% and poured $500 BILLION into fibre optic cables. By 2001, 97% of that fibre sat UNUSED underground. Companies went bankrupt. Trillions of dollars were wiped out. And it’s happening again, but this time at 10 TIMES the scale. So in this case study, I want to answer three questions that no Indian media company is asking. Is this the GREATEST business bet in human history or the BIGGEST bubble ever inflated? Why are the smartest investors on Earth all warning that this is a bubble about to POP? And most importantly, what happens to YOUR job, YOUR investments, and YOUR everyday devices if this bubble bursts?https://lnkd.in/d44VRC_d ### The Dirty AI lie : How the GREATEST bet in human history started to crack in June 2026?#### https://www.youtube.com/ To view or add a comment,sign in
• The best materials scientists—at 10x capacity. Breakthroughs in materials science traditionally require hundreds of scientists, dozens of labs, and decades of work. Radical’s lab invents at the speed of AI, empowering your team to change the trajectory of humankind with advanced materials. To view or add a comment,sign in
• Had fun with this piece recently - our newest paper in Computers in Human Behavior looks at the consequence of exposure to AI generated digital media vs., traditional media. The findings are interesting, and really ask if you are a glass half full, or half empty type of person. Half full - well the influence effect isn’t worse than real images Half empty - so the fake stuff works just as well, and can be created on mass, without any connection to reality and a million times faster So I guess its really just half empty.....https://lnkd.in/gPibf5bU To view or add a comment,sign in

Explore content categories