Timeline lokal

DAMO Academy introduced XuanTie E908A, an automotive-enhanced RISC-V processor design. E908A is a CPU IP core, not a finished chip.
Source: Pandaily
https://pandaily.com/damo-academy-xuantie-e908a-automotive-risc-v-cpu-ip-asil-d-roadmap-c950a
#Semiconductors

US banning Chinese AI DeepSeek from government agency IT systems. DHS told federal agencies to monitor flows of traffic in and out of government websites.
Source: The Hindu
https://www.thehindu.com/sci-tech/technology/us-banning-chinese-ai-deepseek-from-government-agency-it-systems/article71603294.ece
#TechPolicy #DeepSeek

Nscale is negotiating to sell AI computing power to Anthropic in a deal that could be worth about $45 billion.
Source: The Korea Times
https://www.koreatimes.co.kr/world/20261020/nvidia-backed-nscale-anthropic-in-talks-for-new-45-bil-ai-cloud-deal-source-says?utm_source=rss
#AI #Nvidia #Anthropic

A former Google DeepMind research lead and AlphaGo co-creator is launching the London-based AI company Metis Reasoning today. The 12-strong team pursues a compute-efficient alternative to frontier AI models.
Source: Tech.eu
https://tech.eu/2026/10/20/former-google-deepmind-top-researcher-launches-london-compute-efficient-ai-company/
#Startups #Google #DeepMind

SpacemiT introduced the V100 server processor alongside three complete servers, accelerator validation, and operating system support at the RISC-V China Summit.
Source: Pandaily
https://pandaily.com/spacemit-xiangshan-v100-risc-v-server-cpu-c2201-g5201-s5236-kunlunxin-m100
#Semiconductors

Alibaba unveiled an automotive RISC-V processor IP and roadmap for self-driving and robotics.
Source: TechNode
https://technode.com/2026/10/20/alibaba-unveils-automotive-risc-v-processor-ip-and-roadmap-for-self-driving-and-robotics/
#Semiconductors #Alibaba

Chery unveiled its Ziwei automotive chip lineup at its global innovation conference on October 20.
Source: TechNode
https://technode.com/2026/10/20/chery-unveils-automotive-chip-lineup-spanning-ai-computing-silicon-carbide-and-vehicle-control/
#Semiconductors

Tool-using agents are trained with strong simulated users who state every detail clearly, yet real users are often weaker. Researchers propose StaGPO to make weak-user training useful.
Source: arXiv
https://arxiv.org/abs/2610.24294
#MachineLearning

Mirror learns video-to-action mappings through environment interactions without prior action data or hand-crafted rewards.
Source: arXiv
https://arxiv.org/abs/2610.24177
#MachineLearning

A new optimization-grounded framework introduces distributed multi-agent training via ADMM, allowing agents to improve system-level capabilities through decentralized communication.
Source: arXiv
https://arxiv.org/abs/2610.24380
#MachineLearning

TorchTPU is an open-source collaboration between Google and Meta that provides a native PyTorch backend for Google TPUs.
Source: arXiv
https://arxiv.org/abs/2610.24161
#Tech #Google #Meta

AM-Bench is a unified taxonomy and evaluation suite for agentic misalignment. Ten frontier models were evaluated, revealing performance gaps and model-specific behaviors.
Source: arXiv
https://arxiv.org/abs/2610.24264
#MachineLearning

The paper introduces ID-TAP, a neural dueling-bandit framework that personalizes large-scale language model agents at test time by using binary comparisons of candidate responses.
Source: arXiv
https://arxiv.org/abs/2610.24385
#MachineLearning

The paper introduces Software World, a framework that simulates multi-agent coworking in realistic software ecosystems.
Source: arXiv
https://arxiv.org/abs/2610.24061
#MachineLearning

Cortex adds a supervisory layer that governs code generation by a large language model, tracking requirements and iteratively validating and repairing code.
Source: arXiv
https://arxiv.org/abs/2610.24255
#MachineLearning

The paper proposes ACTS, a role-aware adaptive communication topology scheduling framework for efficient large language model multi-agent collaboration.
Source: arXiv cs.MA
https://arxiv.org/abs/2610.25851
#MachineLearning

The paper introduces a one-relation attack that uses a single source-grounded sentence to alter an agent's belief.
Source: arXiv cs.MA
https://arxiv.org/abs/2610.24873
#MachineLearning

Researchers built an AI tool that predicts cancer patients' survival chances using a single photograph of their face.
Source: Club of Mozambique
https://clubofmozambique.com/news/ai-tool-predicts-cancer-survival-chances-with-single-face-photo-study-finds/
#MachineLearning

Saurabh Pandey, founder and CEO of Eloelo, showcased Dolphin AI, an AI video creation tool, at TechSparks, saying it could break the gatekeepers of Indian storytelling.
Source: YourStory
https://yourstory.com/2026/10/eloelos-saurabh-pandey-bets-that-dolphin-ai-can-break-the-gatekeepers-of-indian-storytelling
#Tech

ALS launched a Level 4 autonomous RoboVan for public-road operations and will begin a six-month trial on the 22km Cyberjaya MyAV testing route.
Source: Lowyat.NET
https://www.lowyat.net/2026/407895/als-level-4-autonomous-robovan-trial/
#Robotics