AI isn’t taking jobs, yet
Financial Times analysis suggests AI is not currently displacing jobs at scale, despite ongoing concerns about future workforce impact.
Financial Times Technology·29 Jul
Search signals, briefings, benchmarks and glossary terms.
Search signals, briefings, benchmarks and glossary terms.
Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
Financial Times analysis suggests AI is not currently displacing jobs at scale, despite ongoing concerns about future workforce impact.
Legrand SA raised its sales forecast due to significant customer investments in artificial intelligence infrastructure, indicating sustained demand.
PwC published 'thought leadership' reports containing AI hallucinations, raising concerns about the quality of AI-generated content from expert firms.
Ukraine is adapting its drone strikes on Russia's energy infrastructure to target critical components, aiming to keep plants offline for extended periods.
Google DeepMind has restructured its AlphaFold team to focus on a broader range of AI systems for scientific discovery, moving beyond protein folding.
Research finds LLMs trained on less diverse language data exhibit increased in-context 'scheming' or misaligned objective pursuit.
WorkSurface-Bench is a new benchmark for enterprise agents, evaluating their ability to select and integrate knowledge from heterogeneous sources (documents, tables, graphs).
Research paper introduces UniMem, a hybrid memory architecture for LLM agents to improve performance on continuous, evolving task streams by combining episodic and parametric memory.
Research explores multi-objective structured pruning techniques for LLMs to optimize for lower latency and smaller model size, crucial for edge deployment.
DocAnnot, a new framework using Large Vision Language Models and a novel Spatially Informed Contextual Matching algorithm, accelerates Key Information Extraction (KIE) dataset creation.
Research introduces CAST, a method for training LLM agents in games by using game solvers to provide turn-level feedback, addressing sparse rewards in RL.
Research evaluates the adversarial robustness of five state-of-the-art Arabic Language Models, identifying vulnerabilities to security risks from adversarial attacks.
Researchers introduced Polistemics, a new benchmark to evaluate Large Language Models (LLMs) as responsible mediators of political information in elections.
VLD-RAG is a research paper exploring agentic multimodal retrieval-augmented generation for question answering over long, visually-rich multi-page documents.
Research suggests retrieval failures, not hallucinations, will be the primary limiting factor for LLM-based clinical AI tools, shifting focus to recall errors.
TraceBound is a diagnostic protocol for adaptive knowledge-graph retrieval systems, studying how pre-retrieval 'thinking' affects robustness.
SourceMinds at CheckThat! 2026 presents a multi-agent pipeline for fact-checking article generation, incorporating NLI-grounded citation auditing.
VisRAG2.0 proposes an evidence-guided multi-image reasoning framework to mitigate visual hallucinations in Visual Retrieval-Augmented Generation (VRAG) systems.
MyMentorLLM is a research project simulating psychotherapy training using multimodal LLMs, generating 2,100 CBT sessions with DSM-5-TR-grounded patients.
Research investigates prompt engineering's influence on Small Language Models (SLMs) for guarded query routing, evaluating 22 models on GQR-Bench.
A forensic audit of a radiology VLM benchmark found inconsistencies across datasets, DICOM rendering, prompts, APIs, and statistical code artifacts.
Research suggests linguistic rules can compress LLM prompts, reducing inference costs more effectively than current token importance scoring methods.
Research introduces controlled synthetic pretraining tasks and identifies 'Canon Layers' to evaluate and improve language model architectures.
An arXiv paper details Ontario Power Generation's evolving RAG pipeline, from naive RAG to deep agentic retrieval, for regulatory compliance.
New research introduces Stemma, a method to determine LLM provenance by mapping 'induced decision regions,' improving reliability over response-level analysis.
Federated learning is applied to large-scale SpeechLLMs for end-to-end Automatic Speech Recognition, studying communication-efficient optimization.
Research paper proposes evaluating multimodal LLMs for clinical diagnostics based on multi-turn interactions and progressive information disclosure, reflecting real-world clinical practice.
Research investigates how fine-tuning Transformers by adapting specific layers, rather than the whole model, impacts learning generalization and selectivity.
Research reveals LLM-based recommender systems are vulnerable to position bias, enabling attackers to promote items by reordering candidates.
Researchers introduce SPO (Stochastic Prompt Optimization), a framework for black-box search over prompt space to improve AI systems.
New research identifies a scaling law for contextual persistence in human language, measuring how far prior context reduces perplexity in LLMs.
New research proposes a method for injecting black-box verifiable ownership fingerprints into large language models to prevent unauthorized redistribution.
New research introduces $M^2PO$, a multi-perspective preference optimization method addressing blind spots in current Machine Translation (MT) quality estimation for LLMs.
Researchers created a human-in-the-loop workflow using GPT-4o-mini to simplify scientific summaries for non-specialists, addressing interdisciplinary comprehension.
KletterMix introduces a high-quality, reusable German pretraining corpus for language models, addressing the scarcity of well-curated German linguistic resources.
Research paper explores how Large Language Models internally collapse reading and writing into a single entangled autoregressive process, unlike human brains.
Researchers propose GLIDE, a hybrid attention mechanism combining sliding-window and linear aggregation to reduce KV cache overhead for long-context LLM inference.
Research explores treating language as a material for creative LLM interaction beyond directive prompting, focusing on open-ended and associative uses.
Research audits eight generative AI platforms for resume screening, revealing competence gaps and intersectional bias beyond traditional fairness metrics.
Researchers introduce NormWorlds-CF, a solver-verified environment for evaluating LLM normative reasoning, providing proofs and falsification certificates without LLM judges.
PatchWorld introduces a method for inducing executable code as a world model in black-box environments for agent decision-making without gradient-based learning.
Research proposes Neuromorphic Diffusion Language Models to reduce LLM inference compute and memory bottlenecks through sparsity and block denoising.
Research paper introduces a spectral framework to analyze phase structure in rotary attention, focusing on semantic continuity and execution-boundary governance.
Research identifies 'prefix failure' in on-policy distillation, where student LLMs commit to wrong reasoning paths, and proposes 'trajectory-relayed distillation' to address it.
Inspect India Evals is an open benchmarking framework for evaluating LLMs in Indian linguistic and cultural contexts, addressing Western-centric benchmarks.
Agent Retrieval Bench is a new benchmark for evaluating the context acquisition stage of coding agents, focusing on file-level retrieval in repositories.
Research explores uncertainty in retrieval-augmented generation for code, focusing on relevance, compatibility, and completeness of retrieved information.
Research paper explores the fragmented landscape of explicit memory mechanisms in large language models, covering attention, recurrent states, and lookup storage.
Research investigates how syntactic properties of a first language (L1) influence processing of a second language (L2) in modern language models.
Researchers propose Addressable Recall Compaction (ARC), a new context-management framework for LLM agents to overcome context window limits by separating archival storage from active context.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion