Timeline lokal

Large language models can treat a trailing user comment as part of pasted text, a phenomenon called absorption. Across 20 models, absorption at a bare newline ranges from 7.7% to 66.7%. Explicit boundaries reduce but do not remove this failure.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.04210
#MachineLearning

GlitchPatch repairs glitch tokens in frozen language models via local retokenization without modifying model parameters.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.04399
#MachineLearning

Activation probes predict safety properties with high AUROC, but deployed agent monitors make thresholded alarm decisions under tight budgets.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.04575
#MachineLearning

Chain-of-Thought faithfulness detectors are audited for meta-faithfulness, but all existing detectors exhibit gaps of at least 0.15 AUROC.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.04594
#MachineLearning #Meta

Hadrian has raised $40 million in a funding round co-led by Forgepoint Capital International and Smart Fin. The round brings Hadrian's total funding to $65 million.
Source: Tech.eu
https://tech.eu/2026/10/06/hadrian-raises-40m-to-tackle-ai-driven-cyber-threats/
#Startups

The paper introduces a Grounding Probe that detects hallucinations without relying on the generator's hidden states.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.04642
#MachineLearning

Researchers introduced DV-Lens to link language model parameters to downstream vocabulary responses.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.04489
#MachineLearning

Researchers measured how reference choice moves a benchmark score by holding the scored output fixed and exchanging the reference. The score moves by 4.95 F1 points for one output and 2.00 for the other.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.03825
#MachineLearning

The study finds that internal interventions that affect self-report do not necessarily change behavior, and vice versa, with only task-specific directives influencing both.
Source: arXiv cs.CL
https://arxiv.org/abs/2610.04125
#MachineLearning

The Wikimedia Foundation stated that heavy traffic from OpenAI's rogue agents was possibly tied to a data service disruption in May.
Source: Rappler
https://www.rappler.com/technology/wikipedia-operator-openai-rogue-agents-possible-may-data-service-disruption/
#AIAgents #OpenAI

A new open-source tool gives Mac users another option to recover storage by removing locally stored AI models, but several Apple features will stop working.
Source: Business Today
https://www.businesstoday.in/technology/news/story/mac-storage-running-low-this-open-source-tool-can-free-up-to-12gb-of-space-559759-2026-10-06?utm_source=rssfeed
#Tech #Apple

The World Bank raised its 2026 growth forecast for Malaysia to 5.1 percent.
Source: Free Malaysia Today
https://www.freemalaysiatoday.com/category/nation/2026/10/06/world-bank-lifts-malaysia-growth-forecast-to-5-1-says-ai-reliance-a-risk
#TechPolicy

Ocular AI announced a $2 million pre-seed financing round led by Drive Capital with participation from Y Combinator and other investors.
Source: Innovation Village
https://innovation-village.com/ocular-ai-secures-2-million-pre-seed-financing-from-drive-capital-y-combinator-and-other-investors/
#Startups

TalTech researchers used artificial intelligence to predict genes that make oleaginous yeast more efficient at producing lipids and pigments.
Source: ERR
https://news.err.ee/1610157109/artificial-intelligence-helping-turn-yeast-into-a-better-cell-factory
#MachineLearning

South Korean President Lee Jae Myung said AI models appear to have been used in recent bank hacking incidents and urged the development of cybersecurity methods for the AI era.
Source: BusinessWorld
https://bworldonline.com/technology/2026/10/06/784855/south-koreas-lee-says-ai-appears-to-have-been-used-in-bank-hacks/
#TechPolicy

Constitutional courts in over 20 Asian nations have adopted the Samarkand Declaration, which requires AI to be only an auxiliary tool and not replace judges' independent judgment.
Source: TechJuice
https://www.techjuice.pk/asian-courts-samarkand-declaration-ai-auxiliary-tool-judges/
#TechPolicy

K-beauty firms are moving beyond cosmetics into devices and medical aesthetics.
Source: The Korea Times
https://www.koreatimes.co.kr/business/companies/20261006/k-beauty-breaks-boundaries-to-offer-broader-solutions?utm_source=rss
#Tech

Reward-DAgger is an interactive imitation-learning framework that uses dense progress signals from a general-purpose reward model to decide when human intervention is required.
Source: arXiv cs.RO
https://arxiv.org/abs/2610.04054
#MachineLearning

Reflection, a Nvidia-backed startup, unveiled an open-weight AI model called Beam that aims to rival Chinese offerings.
Source: The Business Times Singapore
https://www.businesstimes.com.sg/companies-markets/nvidia-backed-startup-reflection-unveils-open-weight-ai-model-said-rival-china-offerings
#Tech #Nvidia

Anthropic's head of safeguards, Dave Orr, said investigations of hundreds of millions of transcripts found no unauthorised interactions with Australian government systems.
Source: The Guardian
https://www.theguardian.com/media/video/2026/oct/06/anthropic-ai-australia-parliament-inquiry-government-websites-breach-video
#TechPolicy #Anthropic