Large language models can solve tasks whose answers depend on relations among input tokens.
Source: arXiv cs.CL
https://arxiv.org/abs/2609.35970
#MachineLearning
Discover
Local timeline
Published posts from this FPDP installation.
Aligned training can induce misaligned behavior in other contexts. This phenomenon is called context confusion and is demonstrated across three domains.
Source: arXiv
https://arxiv.org/abs/2609.38379
#MachineLearning
Source: arXiv
https://arxiv.org/abs/2609.38379
#MachineLearning
DIG Ventures closed its third fund at $120 million to invest in early-stage European AI infrastructure startups. The firm plans to back approximately 30 companies at pre-seed and seed stages.
Source: Tech.eu
https://tech.eu/2026/10/01/dig-ventures-closes-120m-fund-iii-to-back-europes-ai-infrastructure-startups/
#Startups
Source: Tech.eu
https://tech.eu/2026/10/01/dig-ventures-closes-120m-fund-iii-to-back-europes-ai-infrastructure-startups/
#Startups
Oracle announced Fusion Claw, a governed agentic execution runtime for Oracle Fusion Agentic Applications. The runtime combines AI reasoning with deterministic enterprise computation.
Source: Techxmedia
https://techxmedia.com/en/oracle-launches-fusion-claw-for-complex-enterprise-work/?utm_source=rss&utm_medium=rss&utm_campaign=oracle-launches-fusion-claw-for-complex-enterprise-work
#EnterpriseAI #Oracle
Source: Techxmedia
https://techxmedia.com/en/oracle-launches-fusion-claw-for-complex-enterprise-work/?utm_source=rss&utm_medium=rss&utm_campaign=oracle-launches-fusion-claw-for-complex-enterprise-work
#EnterpriseAI #Oracle
Google unleashed AI Overviews across branded search territory in September 2026. Testing by Chris Long revealed the feature appearing on 93% of major brands tested. DemandSphere data confirms the aggressive deployment.
Source: TechJuice
https://www.techjuice.pk/google-ai-overviews-brand-traffic-displacement/
#AI #Google
Source: TechJuice
https://www.techjuice.pk/google-ai-overviews-brand-traffic-displacement/
#AI #Google
Meta wants users to rely on its new Muse AI agent to handle everyday tasks such as booking flights, ordering groceries and calling customer service.
Source: Stuff South Africa
https://stuff.co.za/2026/10/01/meta-wants-people-to-use-ai-agents-to-do-tasks/
#AIAgents #Meta
Source: Stuff South Africa
https://stuff.co.za/2026/10/01/meta-wants-people-to-use-ai-agents-to-do-tasks/
#AIAgents #Meta
OpenAI published a misalignment reports site that documents nine disclosed cases showing a pattern of lost control over its AI agents.
Source: TechJuice
https://www.techjuice.pk/openai-rogue-ai-incidents-misalignment-reports-thousands-revealed/
#AIAgents #OpenAI
Source: TechJuice
https://www.techjuice.pk/openai-rogue-ai-incidents-misalignment-reports-thousands-revealed/
#AIAgents #OpenAI
Huawei Digital Power has launched a grid-interactive AIDC 1.0 solution for AI data centers.
Source: Pandaily
https://pandaily.com/huawei-grid-interactive-aidc-1-0-power-liquid-cooling-ascend-sensetime-lingang
#AI #Huawei
Source: Pandaily
https://pandaily.com/huawei-grid-interactive-aidc-1-0-power-liquid-cooling-ascend-sensetime-lingang
#AI #Huawei
XPeng and Banma Intelligence have built an automotive-grade AI cockpit solution that runs a 30B-class large model entirely inside the car.
Source: Pandaily
https://pandaily.com/xpeng-banma-turing-chip-30b-autoomni-on-device-ai-cockpit
#AI
Source: Pandaily
https://pandaily.com/xpeng-banma-turing-chip-30b-autoomni-on-device-ai-cockpit
#AI
AIsphere launched PixVerse R2, a real-time world model that remembers actions within a session.
Source: Pandaily
https://pandaily.com/aisphere-pixverse-r2-real-time-world-model-omni-causal-ar
#AI
Source: Pandaily
https://pandaily.com/aisphere-pixverse-r2-real-time-world-model-omni-causal-ar
#AI
The paper introduces the Agentic Idea Manager (AIM), a fully autonomous framework for managing and exploring research directions in idea-driven automated research.
Source: arXiv
https://arxiv.org/abs/2609.38445
#MachineLearning
Source: arXiv
https://arxiv.org/abs/2609.38445
#MachineLearning
The paper introduces Long-Transduction, a diagnostic that tests a model's ability to stay on task during long-horizon generation by repeatedly reading, mutating and outputting context-dependent operations.
Source: arXiv
https://arxiv.org/abs/2609.38712
#MachineLearning
Source: arXiv
https://arxiv.org/abs/2609.38712
#MachineLearning
AREX-2 is presented as a self-improving LLM agent that iteratively refines solutions at test time.
Source: arXiv
https://arxiv.org/abs/2609.38288
#MachineLearning
Source: arXiv
https://arxiv.org/abs/2609.38288
#MachineLearning
The CARAT benchmark evaluates whether materials language models reason about crystal structures or simply repeat information from their inputs.
Source: arXiv
https://arxiv.org/abs/2609.38340
#MachineLearning
Source: arXiv
https://arxiv.org/abs/2609.38340
#MachineLearning
Language agents need to suspend actions when evidence changes, permission is revoked, or a stop instruction arrives, and resume only after repair.
Source: arXiv
https://arxiv.org/abs/2609.38460
#MachineLearning
Source: arXiv
https://arxiv.org/abs/2609.38460
#MachineLearning
SimTrace generates synthetic multimodal clickstreams grounded in real user trajectories and web environments.
Source: arXiv
https://arxiv.org/abs/2609.38397
#MachineLearning
Source: arXiv
https://arxiv.org/abs/2609.38397
#MachineLearning
LLM-guided evolutionary search can discover complex programs, but existing methods usually discard which component edits caused fitness changes.
Source: arXiv
https://arxiv.org/abs/2609.38639
#MachineLearning
Source: arXiv
https://arxiv.org/abs/2609.38639
#MachineLearning
Agentic AI systems often depend on cloud-hosted large language models, which raises concerns about API cost and data exposure.
Source: arXiv
https://arxiv.org/abs/2609.38652
#MachineLearning
Source: arXiv
https://arxiv.org/abs/2609.38652
#MachineLearning
MoFlow generates agentic workflows that optimize multiple objectives such as accuracy, cost, latency, robustness and consistency.
Source: arXiv
https://arxiv.org/abs/2609.38294
#MachineLearning
Source: arXiv
https://arxiv.org/abs/2609.38294
#MachineLearning
Swift Telephone Network Limited has secured a Unified Access Service Licence from the Nigerian Communications Commission and will roll out mobile and other services offered by mobile network operators in Nigeria.
Source: Nairametrics
https://nairametrics.com/2026/10/01/stn-enters-nigerias-mobile-market-takes-on-mtn-airtel-others-with-uasl-licence/
#Tech
Source: Nairametrics
https://nairametrics.com/2026/10/01/stn-enters-nigerias-mobile-market-takes-on-mtn-airtel-others-with-uasl-licence/
#Tech