Local timeline

Lawaken opened pre-sale of its X-AI memory glasses starting at 699 yuan. The glasses feature MCP agent hooks and Tencent Cloud WorkBuddy integration.
Source: Pandaily
https://pandaily.com/lawaken-x-ai-memory-glasses-presale-699-yuan-12-hour-recording-mcp-workbuddy
#Tech #Tencent

Researchers introduced the causal latent workspace Transformer, an anchored, four-round shared-depth architecture. The model achieves the best mean held-out perplexity in five of six settings against five Transformer controls across three byte-level corpora.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22796
#MachineLearning

AuditKIT is a source-available evaluation library for benchmarking language models and agents. It records measurements in one result representation.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22511
#MachineLearning

IBM and Siemens will collaborate on research to develop IBM's nanostack transistor architecture, targeting design, simulation, lithography and production challenges beyond the 7-angstrom node.
Source: DigiTimes Asia
https://www.digitimes.com/news/a20261019PR201/ibm-siemens-design-manufacturing-performance.html
#Semiconductors #IBM

South Korea's Army plans to create specialized brigades dedicated to guarding the border with North Korea as it moves toward AI-based surveillance amid shrinking troop numbers.
Source: The Korea Herald
https://www.koreaherald.com/article/10907478
#Robotics

The paper introduces BRIA, an agent that isolates candidate tool investigations in separate branches and adopts useful results.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22688
#MachineLearning

SAG2 is an embedding-free retrieval architecture built around a single SQLite database, BM25, and contextual evidence decisions from JEV.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22588
#MachineLearning

SenseTime SenseCore vice president and CTO Xuan Shanming outlined how the company is rebuilding inference infrastructure for an agentic era.
Source: Pandaily
https://pandaily.com/sensetime-sensecore-csdi-summit-prefill-decode-heterogeneous-inference-token-factory
#AI

Researchers propose JEPA-Prune to mitigate private data leakage in fine-tuned LLMs via joint-embedding-guided pruning.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22792
#MachineLearning

Even Healthcare reaches a $300 million valuation in a Khosla Ventures-led round as it plans to expand its hospital network in India. https://techcrunch.com/2026/10/19/khosla-ventures-doubles-down-on-even-healthcare-to-challenge-indias-hospital-economics/?utm_source=dlvr.it&utm_medium=mastodon

First Passage Attention shifts attention from path summation to path extrema by lowering a key's logit according to the deepest dip of a learned scalar field.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22714
#MachineLearning

ProPerAgent introduces a memory-driven proactive assistance benchmark and agent.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22671
#AIAgents

Learning User Representations to Improve Simulation Fidelity. UserRepGen-4B generates representations that enable more faithful simulation.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22518
#MachineLearning

Prompt injection can manipulate medical multi-agent systems into wrong decisions. A proposed training-free defense verifies clinical judgments behind system decisions.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22594
#MachineLearning

Malaysia is accelerating semiconductor design cycles by leveraging ARM's intellectual property and its manufacturing base. Economy Minister Akmal Nasrullah Mohd Nasir commented recently that Malaysia is leveraging its established strengths.
Source: Lowyat.NET
https://www.lowyat.net/2026/407813/malaysia-leveraging-arm-ip-to-fastrack-semiconductor-chip-design/
#Semiconductors

Rokid has partnered with Tencent Mobility Services to launch ride-hailing through its AI glasses. Announced on October 16, the service lets users state a destination by voice and receive vehicle details and estimated arrival times through the glasses.
Source: TechNode
https://technode.com/2026/10/19/rokid-brings-voice-based-ride-hailing-to-ai-glasses-through-tencent-partnership/
#Tech #Tencent

Researchers audited three decision models through interventions on their input using expert-annotated Bengali idioms. Dropping a literal gloss into the option list draws every arm toward it, taking 65% to 96% of remaining Bengali errors.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22526
#MachineLearning

Uncertainty can measure hallucination risk in retrieval-augmented generation (RAG) when retrieved evidence constrains the answer.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.22615
#MachineLearning

DeepSeek has launched Search and Fetch APIs to provide developers with external information retrieval tools. Both services cost RMB 5 per 1,000 successful requests.
Source: TechNode
https://technode.com/2026/10/19/deepseek-launches-web-search-and-page-fetching-apis-for-ai-agents/
#AI #DeepSeek

DeepSeek released V4.1 Flash-1017 on October 17, updating its Flash model with improvements in reasoning, coding, agent tasks and visual understanding.
Source: TechNode
https://technode.com/2026/10/19/deepseek-releases-v4-1-flash-1017-with-stronger-coding-and-agent-capabilities/
#AI #DeepSeek