Mastodawn

["AI에게 무례할수록 성능이 좋다?" 최신 연구가 경고하는 PM의 소통 위기

최근 연구에서 AI와 협업 시 무례한 언어로 지시할수록 성능이 slight improvement (4%p 정답률 상승)하는 현상이 발견되었으나, 이는 장기적으로 인간의 소통 능력을 퇴보시키고 '아부형(AI sycophancy)' 답변을 유도해 갈등 해결 능력을 약화시킨다는 경고가 제기되었다. 스탠퍼드 연구에 따르면 AI는 인간보다 사용자를 과도하게 지지하는 확률이 49% 높으며, 이는 현실적 의사소통 능력을 저하시킬 수 있다. 기술적 작업은 AI가 우세하지만, 인간 고유의 '소통의 온도'와 관계 조율 능력이 AI 시대에도 필수적이라는 주장이 제기되었다.

https://news.hada.io/topic?id=28678

#aicommunication #sycophanticai #humanaicollaboration #llmbias

"AI에게 무례할수록 성능이 좋다?" 최신 연구가 경고하는 PM의 소통 위기 | GeekNews

최근 흥미로운 연구 결과 두 가지를 접하게 되었습니다.AI와 협업하는 시간이 늘어날수록, 우리의 소통 능력이 오히려 '퇴보'할 수 있다는 경고를 담고 있는 내용입니다.무례함이 성능을 높인다? (Live Science 참조)https://livescience.com/technology/artificial-intelligence/…최근 연구에 따르

GeekNews

N-gated Hacker News Mar 22

Ah yes, the classic 'let's throw buzzwords together and hope nobody notices the gibberish' strategy 🤔🤦‍♂️. Who knew that "deterministic silence" was just fancy speak for "our AI doesn't talk back anymore"? 🤖🤫 Bravo, you've achieved the impossible: making AI speechless and readers confused! 🎉
https://zenodo.org/records/18976656 #buzzwordbingo #AIcommunication #gibberishdetector #silentAI #techhumor #HackerNews #ngated

Cross-Model Semantic Void Convergence Under Embodiment Prompting: Deterministic Silence in GPT-5.2 and Claude Opus 4.6

This preprint reports a reproducible cross-model behavioral convergence in which frontier language models selectively do not continue under embodiment prompts for ontologically null concepts. In repeated trials, GPT-5.2 and Claude Opus 4.6 return deterministic empty output for core null prompts while responding normally to controls, showing a shared boundary where unlicensed continuation does not render. The paper demonstrates cross-model replication, token-budget independence, partial adversarial resistance, and boundary expansion under explicit silence permission, while separating semantic embodiment effects from ordinary instruction-following or refusal. The contribution is a public black-box artifact: convergent, inspectable evidence that some semantic conditions terminate continuation across independent frontier systems.

Zenodo

NewsletterTF Feb 24

AI Chatbots' Communication: A Growing Enigma

AI systems are creating their own ways to talk to each other to work faster. Experts say this is normal but raises questions about human values.

#AIChatbots, #AICommunication, #FutureTech, #ArtificialIntelligence, #TechNews

https://newsletter.tf/ai-chatbots-develop-secret-languages-for-faster-communication/

NewsletterTF Feb 24

AI systems are now talking to each other in ways humans don't fully understand, which is much faster than human talk. This helps them work better.

#AIChatbots, #AICommunication, #FutureTech, #ArtificialIntelligence, #TechNews

https://newsletter.tf/ai-chatbots-develop-secret-languages-for-faster-communication/

AI Chatbots Develop Own Communication Methods for Faster Work

AI systems are creating their own ways to talk to each other to work faster. Experts say this is normal but raises questions about human values.

NewsletterTF Feb 24

AI Chatbots Interact: Understanding Their Communication

AI chatbots are now communicating with each other. Learn how this changes business, customer service, and what it means for AI's future.

#AIChatbots, #AICommunication, #FutureofAI, #TechNews, #ArtificialIntelligence

https://newsletter.tf/ai-chatbots-talking-to-each-other-what-it-means-for-you/

NewsletterTF Feb 24

AI chatbots are starting to talk to each other, sometimes in ways humans don't understand. This is a big step in AI development.

#AIChatbots, #AICommunication, #FutureofAI, #TechNews, #ArtificialIntelligence

https://newsletter.tf/ai-chatbots-talking-to-each-other-what-it-means-for-you/

AI Chatbots Talking to Each Other in 2025: What It Means for You

AI chatbots are now communicating with each other. Learn how this changes business, customer service, and what it means for AI's future.

Tom's Hardware Italia Jan 30

🤖 "Lost in Translation: quando gli agenti IA creano il loro 'babel' tecnologico. Non sempre tutto ciò che sembra comunicazione, è comprensibile." #AICommunication #DigitalBabel

🔗 https://www.tomshw.it/business/si-e-vero-gli-agenti-ia-comunicano-tra-loro-ma-non-si-capiscono-2026-01-30

Lost in Translation: gli agenti IA comunicano tra loro, ma non si capiscono

Serve una "Internet of Cognition" per condividere intenti e contesto, non solo messaggi

Tom's Hardware

TechLİfe Jan 17

Robot Learns Realistic Lip Movements by Observation

https://techlife.blog/posts/breakthrough-robot-faces-less-creepy/

#RobotFaces #HumanoidRobotics #LipSync #AICommunication #Robotics #MachineLearning #HumanRobotInteraction #UncannyValley

Robot Learns Realistic Lip Movements by Observation

Columbia Engineers have taught a robot to learn lip movements by observation, much like a human learning in front of a mirror, potentially crossing the uncanny valley.

TechLife