Timeline lokal

Beijing has launched a centrally managed state-owned enterprise to create a national data group aimed at turning the country's data into fuel for artificial intelligence.
Source: South China Morning Post
https://www.scmp.com/tech/policy/article/3372159/chinas-ai-race-needs-data-fuel-can-new-state-owned-giant-deliver
#TechPolicy

Google updated its privacy policy on October 16 to allow ad personalization using data from Gmail, Drive and Photos for users who enable the Personal Intelligence feature.
Source: MediaNama
https://www.medianama.com/2026/10/223-google-gmail-data-ad-personalisation/
#TechPolicy #Google

Union Minister Ashwini Vaishnaw stated that a consultation paper on draft AI regulations will be issued in the coming weeks. It remains unclear whether a new law or the existing IT Act will be used.
Source: MediaNama
https://www.medianama.com/2026/10/223-india-ai-regulation-consultation-paper-vaishnaw/
#TechPolicy

Elvis Presley's music owners have sued AI firm Anthropic, alleging it used 274 of his songs to train its Claude models without permission.
Source: Gulf News
https://gulfnews.com/entertainment/elviss-rights-holders-seek-411-million-from-anthropic-over-songs-used-to-train-ai-1.500723735
#TechPolicy #Anthropic #Claude

The World Bank launched its first digital strategy on 20 October 2026 in Seoul, linking affordable connectivity to job creation and drawing private capital into developing markets.
Source: Further Africa
https://furtherafrica.com/2026/10/26/world-bank-digital-strategy-targets-jobs-and-access/
#TechPolicy

UK-based Nous has raised $17 million in new funding to expand internationally and introduce additional features.
Source: Tech.eu
https://tech.eu/2026/10/26/nous-lands-17m-to-bring-ai-powered-household-management-to-more-homes/
#Startups

ByteDance's Seedance 2.0 Pro took top spot in a ranking of 12 leading artificial intelligence video-generation models, scoring 3.84 out of 5 for visual quality.
Source: South China Morning Post
https://www.scmp.com/tech/tech-trends/article/3372107/chinese-video-ai-models-beat-us-rivals-visual-quality-lag-safeguards-hku-study
#AI #ByteDance

The eBeam Initiative conducted its 15th annual survey and reported the results at its annual reception.
Source: Semiconductor Engineering
https://semiengineering.com/large-format-masks-move-from-aspiration-to-industry-roadmap/
#Semiconductors

Researchers introduced EDWARD-NMT, a novel MoE encoder-decoder architecture designed for a train once, deploy anywhere paradigm.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.32115
#MachineLearning

Researchers propose DIAL-S2ST, a post-training method for learning inference-time latency control in simultaneous speech-to-speech translation models.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.32214
#MachineLearning

An innovation team incubated by Xiaomi's technology committee has shown HachimoDock. The device appeared at the 2026 Guancha Conference on October 26.
Source: Pandaily
https://pandaily.com/xiaomi-hachimodock-open-source-desktop-ai-device-voice-agents-codex-workbuddy-miloco
#AIAgents

Large language models often exhibit sycophancy, altering their answers to align with user suggestions even on objective factual and reasoning tasks.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.32055
#MachineLearning

DraMaze is a multi-agent pipeline proposed for diverse microdrama script generation.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.32002
#MachineLearning

Researchers at imec published Beyond HBM-on-GPU: Thermal Design Envelope for 3D Volumetric DRAM-on-GPU Integration. They evaluated a 3D memory architecture that rotates DRAM dies vertically above a GPU.
Source: Semiconductor Engineering
https://semiengineering.com/vertical-dram-stacking-expands-thermal-design-space-beyond-hbm-imec/
#AI

When a language model is asked whether its own answer is correct, three languages are involved: the question language, the response language, and the verifier language.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.32144
#MachineLearning

TaxoSAE integrates taxonomic information into sparse autoencoders. arXiv cs.CL
Source: arXiv cs.CL
https://arxiv.org/abs/2610.31894
#MachineLearning

An external model trained on a policy's behavioral traces can estimate confidence as accurately as the policy itself. For LLMs trained with value-based RL, the existing critic can be reused as an off-the-shelf confidence estimator.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.31820
#MachineLearning

The Schema Coherence Inventory (SCI) is a 96-item forced-choice test that scores language model outputs by length-normalized log-likelihood without generating or parsing text.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.31683
#MachineLearning

India's voice AI challenges could shape global technology, according to Murf AI co-founder Sneha Roy.
Source: YourStory
https://yourstory.com/2026/10/india-voice-ai-murf-ai-sneha-roy-techsparks
#Tech

Large language models can achieve distributional alignment and generalize to unseen questions of the same domain, but generalization can be unpredictable and unstable.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.31687
#MachineLearning