Timeline lokal

Taiwan's first semiconductor strike ended after a 24-hour warning action on October 23, 2026.
Source: DigiTimes Asia
https://www.digitimes.com/news/a20261023PD239/strike-taiwan-micron-management-2026.html
#Semiconductors #Micron

Apple has officially entered the foldable smartphone market with the launch of the iPhone Duo. Sales of the device began on Friday, marking Apple's first direct challenge to established foldable manufacturers.
Source: Innovation Village
https://innovation-village.com/apples-iphone-duo-arrives-can-it-do-for-foldables-what-the-iphone-did-for-smartphones/
#Tech #Apple

ASUS unveiled its Enterprise Agentic AI Solution in Berlin, introducing an integrated full-stack offering.
Source: Vietnam Investment Review
https://vir.com.vn/asus-unveils-enterprise-agentic-artificial-intelligence-servers-in-berlin-163450.html
#AI

Barclays calls AI 'icing on the cake' as it culls jobs. The company expects over R24 billion in savings in 2027.
Source: ITWeb
https://www.itweb.co.za/article/barclays-calls-ai-icing-on-the-cake-as-it-culls-jobs/Kjlyr7wBQm6vk6am
#EnterpriseAI

Group Relative Policy Optimisation improves multilingual speech recognition in low-resource languages. SunflowerASR uses Whisper-large-v3 across 51 African languages.
Source: arXiv
https://arxiv.org/abs/2610.30663
#MachineLearning

AgentDP automates model optimization and adaptation for heterogeneous embedded AI applications.
Source: arXiv
https://arxiv.org/abs/2610.30760
#MachineLearning

Researchers propose DSWM, a Distributional Social World Model predicting alternative responses, probabilities, and outcome values in one output per action.
Source: arXiv
https://arxiv.org/abs/2610.30657
#MachineLearning

Personalized GUI agents must adapt execution based on user preferences, but existing benchmarks only measure overall task success.
Source: arXiv
https://arxiv.org/abs/2610.30573
#MachineLearning

A new study presents a full-length measurement of acceptance rate as a function of generation position, finding that the acceptance rate does not decay with position.
Source: arXiv
https://arxiv.org/abs/2610.30644
#MachineLearning

A judge in New Mexico ordered Meta to pay a fine of about $25 billion after finding it guilty of misleading Facebook users about its data privacy policies.
Source: EL PAÍS English
https://english.elpais.com/technology/2026-10-23/metas-annus-horribilis-why-it-was-hit-with-the-largest-fine-ever-imposed-on-a-tech-company.html
#TechPolicy #Meta

Lenovo Chairman and CEO Yuanqing Yang defined a new mission to bring artificial intelligence to every home and industry while targeting over US$100 billion in annual revenue within two years.
Source: DigiTimes Asia
https://www.digitimes.com/news/a20261023PD225/lenovo-nvidia-infrastructure-rubin-manufacturing.html
#Tech #Nvidia

Raw-RAG ranks first in six of seven benchmark-model settings, outperforming the strongest agentic memory system.
Source: arXiv
https://arxiv.org/abs/2610.30755
#MachineLearning

Researchers ask whether large language models can serve as event world models for systems with multiple actors. Models struggle when other actors strongly affect outcomes.
Source: arXiv
https://arxiv.org/abs/2610.30488
#MachineLearning

CLI provides superior efficiency for single-user workflows, while MCP addresses requirements that CLI fundamentally cannot satisfy.
Source: arXiv
https://arxiv.org/abs/2610.30157
#MachineLearning

An empirical study of handshake overhead in AI agent communication protocols presents phase-decomposed benchmarks for MCP, A2A, and ACP. A legacy MCP handshake takes three round trips and 230 ms at 50 ms RTT.
Source: arXiv
https://arxiv.org/abs/2610.30151
#MachineLearning

Researchers addressed the high-degree regime in Lehmer's problem using a human-AI discovery loop. They derived a five-parameter construction called the Generalized Boyd algorithm.
Source: arXiv
https://arxiv.org/abs/2610.30277
#MachineLearning

Faithful latent harm monitors work no better than models own behaviour at a matched false alarm rate.
Source: arXiv
https://arxiv.org/abs/2610.30727
#MachineLearning

Microsoft developed an AI diagnostic tool that solved complex medical cases more than twice as accurately as generalist doctors in a study published in Science.
Source: The Next Web
https://thenextweb.com/news/microsoft-ai-diagnosis-mai-dxo-science-study
#AI #Microsoft

Researchers trained a reinforcement learning policy in a self-judged tournament to improve moral argument generation in language models.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.30349
#MachineLearning

Japan will create a team of ethical hackers that will use artificial intelligence to test the nation's own systems for vulnerabilities.
Source: The Next Web
https://thenextweb.com/news/japan-ai-hacker-team-php-flaws-softbank
#TechPolicy