Timeline lokal

Researchers released StreamDecisionBench, a dataset of eight streaming scenarios in four application families.
Source: arXiv cs.CL
https://arxiv.org/abs/2609.38612
#MachineLearning

The paper introduces STAM, a state-transition-aware memory framework that records state changes as new clinical entries arrive.
Source: arXiv cs.CL
https://arxiv.org/abs/2609.38490
#MachineLearning

Researchers examined how system prompts alter language model computation across 17 instruction-tuned models using Centered Kernel Alignment. Safety instructions barely move intermediate representations compared to persona instructions.
Source: arXiv cs.CL
https://arxiv.org/abs/2609.38205
#MachineLearning

A new telecom operator named STN has entered a market dominated by MTN and Airtel. The UASL authorises STN to provide voice, data, fixed and mobile telecommunications services.
Source: TechEconomy Nigeria
https://techeconomy.ng/new-telecom-operator-stn-enters-market-dominated-by-mtn-airtel

Research characterizes PRM-Pruned Fragment Grafting as statistically indistinguishable from an independent parallel-CoT baseline.
Source: arXiv
https://arxiv.org/abs/2610.00047
#MachineLearning

The audit of text-to-3D evaluation used 300 frozen scenes from six generators and varied eight rendering and caption factors.
Source: arXiv
https://arxiv.org/abs/2610.00447
#MachineLearning

Agent evaluations are increasingly used to compare LLMs, yet ranks can reflect evaluation conditions.
Source: arXiv
https://arxiv.org/abs/2610.00651
#MachineLearning

A new method purifies LoRA-tuned LLMs via null-space projection without requiring trigger knowledge or retraining.
Source: arXiv
https://arxiv.org/abs/2610.00685
#MachineLearning

Lithuania is testing a Ukrainian-developed airspace monitoring system that detects low-flying drones and other threats by analysing acoustic and radio signals, the defence ministry said on Friday.
Source: LRT English Lithuania
https://www.lrt.lt/en/news-in-english/19/3070136/lithuania-tests-ukrainian-developed-drone-detection-system-skyfortress
#Robotics

Role-specialized QA pipelines pass rationales from a reasoner to a verifier. Faithful rationales add almost no answer accuracy over no rationale. Corrupted rationales strongly alter support judgments.
Source: arXiv
https://arxiv.org/abs/2610.00018
#MachineLearning

Micron is expanding HBM capacity across four major sites in Taiwan. Donghui Lu stated that AI is driving strong memory demand.
Source: DigiTimes Asia
https://www.digitimes.com/news/a20261002PD224/micron-taiwan-capacity-hbm-2026.html
#Semiconductors #Micron

London healthtech startup Lottie has acquired CareMaster, its second care software acquisition. The acquisition expands its use of AI in care operations.
Source: Tech.eu
https://tech.eu/2026/10/02/lottie-snaps-up-caremaster-as-ai-push-moves-deeper-into-care-operations/
#Startups

Zhou Yuxiang hired food delivery riders from Meituan and Ele.me as salespeople for Black Lake Technologies after traditional software professionals failed to access factories.
Source: Fortune
https://fortune.com/2026/10/02/black-lake-yuxiang-zhou-ai-china-google-moment/
#EnterpriseAI

Edward Tian built GPTZero over winter break of his senior year at Princeton. Less than four years later, GPTZero has more than 20 million registered users. Superhuman agreed in June to acquire GPTZero for undisclosed terms.
Source: Fortune
https://fortune.com/2026/10/02/hallucination-detection-gptzero-edward-tian-alex-cui/
#Startups

GAVA implements grounded correction arbitration with observation-bounded evidence, legal probes, and a one-step expected-loss rule.
Source: arXiv
https://arxiv.org/abs/2610.00282
#MachineLearning

Researchers frame LLM agent recovery as a causal decision problem to balance rescue and harm. The Causal Intervention Router improves long-horizon ALFWorld success from 70.33% to 73.33%.
Source: arXiv
https://arxiv.org/abs/2610.00372
#MachineLearning

China has set a five-year goal for the expansion of intelligent computing capacity to drive artificial intelligence that calls for a more than fourfold increase on what was in place at the end of June this year.
Source: South China Morning Post
https://www.scmp.com/plus/tech/policy/article/3369537/chinas-ai-compute-ambitions-demand-innovation?utm_source=rss_feed
#AI

Researchers introduced a chess benchmark built from 1,118 Lichess puzzles to study automatic prompt optimization for frozen large language models.
Source: arXiv
https://arxiv.org/abs/2610.00416
#MachineLearning

JevSpawn is a compositional policy that links natural language task specifications to finite probabilistic exploration.
Source: arXiv
https://arxiv.org/abs/2610.00437
#MachineLearning

Researchers study heavy-tailed memory traces in long-horizon language agents and propose the Core--Tail World Model controller.
Source: arXiv
https://arxiv.org/abs/2610.00010
#MachineLearning