Timeline lokal

Investors are still betting on AI, but doubts are piling up as the AI consensus starts to crack.
Source: Fortune
https://fortune.com/2026/10/22/ai-consensus-crack-investors-doubts-anthropic-ipo/
#Startups

Researchers introduce LTEM, an LLM-friendly Table-Enhanced Markdown designed for table understanding and generation.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.29188
#MachineLearning

Speculative decoding uses a target model to verify a tree of draft tokens in parallel. We study whether the resulting next-token distributions can help select the next tree.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.29210
#MachineLearning

The paper introduces Role-Enhanced Zero-Shot Chain-of-Thought prompting, which adds domain-specific expert roles to improve reasoning in large language models.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.29073
#MachineLearning

Researchers present Lift-Bench and Lift-Train to address reward hacking in language models. Common instruction following benchmarks may lead to reward hacking.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.28819
#MachineLearning

Signals and Queries is a token-efficient method for discovering relevant policies in user-facing LLM agents.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.29326
#MachineLearning

TrainBridge is an agentic approach for transferring GPU optimization guidance to NPU training.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.29171
#MachineLearning

Current LLM safety benchmarks predominantly rely on binary safe or unsafe labels. We introduce DiagSafe, a diagnostic evaluation framework that moves beyond binary classification by employing prompt-level rubrics.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.29225
#MachineLearning

SkillEval is a framework that uses large language models to discover interpretable skills and cognitive diagnosis models to estimate per-skill abilities.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.28767
#MachineLearning

Researchers introduced the Hierarchical Masked Diffusion Language Model to incorporate priority among token positions by augmenting each position with a hierarchy level.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.29255
#MachineLearning

Researchers study multi-turn conversations where two LLMs discuss subjective matters, can agree or disagree, and may change their stances.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.29328
#MachineLearning

Boomi CEO Steve Lucas stated that artificial intelligence now reviews half of the integration company's accounts payable. Lucas also noted that Boomi is approaching $1bn in recurring revenue and has tripled its revenue over the past four years.
Source: The Next Web
https://thenextweb.com/news/steve-lucas-boomi-ai-accounts-payable-tnw-podcast
#Tech

Millions of hacking attempts have triggered a global cybersecurity alert as entities race to patch critical PHP vulnerabilities.
Source: Cyprus Mail
https://cyprus-mail.com/2026/10/22/millions-of-hacking-attempts-trigger-global-cybersecurity-alert
#Tech

NACSA urges local organisations to immediately patch internet-facing PHP systems following the discovery of two critical vulnerabilities known as PHP2Shell.
Source: Lowyat.NET
https://www.lowyat.net/2026/408157/nacsa-urges-php-systems-patch/
#TechPolicy

Harbor now works with Docker Sandboxes, letting each evaluation task run in its own microVM instead of sharing a container.
Source: InfoQ
https://www.infoq.com/news/2026/10/ai-agents-harbor-docker/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global
#AIAgents

Schneider Electric, AVEVA, Nscale and Start Campus announced an Immersive Twin pilot built with the NVIDIA Omniverse DSX Blueprint. The pilot connects Schneider Electric, AVEVA and ETAP technology in one digital platform.
Source: Techxmedia
https://techxmedia.com/en/immersive-twin-pilot-brings-digital-twins-to-ai-factories/?utm_source=rss&utm_medium=rss&utm_campaign=immersive-twin-pilot-brings-digital-twins-to-ai-factories
#AI #Nvidia

Semantic leakage occurs when an irrelevant concept in the prompt steers model generation about an entity. Research shows that leakage persists even after removing the attribution of the concept to the specific entity.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.28944
#MachineLearning

Barclays is cutting 5,000 jobs, about 5% of its 93,000 global staff, as its bosses cite advances in artificial intelligence.
Source: The Guardian
https://www.theguardian.com/business/2026/oct/22/barclays-profits-up-as-uk-banks-brace-for-tax-increase-in-budget
#Automation

Anthropic released the Claude Duet AI Assistant Research Preview as a standalone iOS and Android app. Powered by the Opus 5.5 model, the assistant supports over 6,000 connectors.
Source: Gadgets 360
https://www.gadgets360.com/ai/news/claude-duet-ai-assistant-research-preview-app-rolled-out-opus-5-5-model-12219233#rss-gadgets-all
#AI #Claude #Anthropic

A New Mexico judge ordered Meta Platforms to pay $25 billion after a jury found the company made misleading claims about Facebook user privacy.
Source: BusinessDay
https://businessday.ng/technology/article/meta-ordered-to-pay-25bn-over-misleading-claims-about-facebook-user-privacy/
#TechPolicy #Meta