The US Department of Commerce is expanding its AI safety testing: Following Anthropic and OpenAI, Google Deepmind, Microsoft, and xAI have now signed agreements with the Center for AI Standards and Innovation. The companies provide models with reduced safety guardrails f
OpenAI is swapping out ChatGPT's default model for GPT-5.5 Instant. In internal testing, the update produced 52.5 percent fewer hallucinated claims on high-risk topics like medicine and law. A new feature called "memory sources" lets users see which stored context s
arXiv:2605.02712v1 Announce Type: cross Abstract: SemEval-2026 Task 10 is focused on conspiracy detection. Specifically, the goal is to detect whether a Reddit comment expresses a conspiracy belief. Our submitted mdok-style system utilizes data augmentation and self-training (to cope with a rather s
arXiv:2507.15143v3 Announce Type: replace Abstract: This paper investigates the feasibility of human mobility in extreme urban morphologies characterized by high-density vertical structures and linear city layouts. To assess whether agents can navigate efficiently within such unprecedented topologie
arXiv:2512.04032v3 Announce Type: replace-cross Abstract: We present jina-vlm, a token-efficient 2.4B parameter vision-language model that achieves state-of-the-art multilingual VQA performance among open 2B-scale VLMs. The model couples a SigLIP2 vision encoder with a Qwen3 language decoder and mak
arXiv:2511.21086v2 Announce Type: replace Abstract: Large language models must satisfy hard orthographic constraints during controlled text generation, yet systematic cross-family evaluation remains limited. We evaluate 39 configurations spanning three model families (Qwen3, Claude Haiku 4.5, GPT-5-
BLF (Bayesian Linguistic Forecaster) achieves state-of-the-art performance on ForecastBench by combining numerical probability estimates with natural-language evidence summaries in a linguistic belief state. This agentic system uses iterative tool-use loops to refine forecasts through structured reasoning.
Research investigating value alignment robustness through finetuning using animal compassion as an orthogonal alignment dimension. Introduces the Animal Harm Benchmark (AHB) with 26 questions across 13 ethical dimensions for evaluating compassionate reasoning in LLMs.
MetaErr addresses the unpredictability of deep learning failures by developing methods to predict when neural networks will fail. While deep learning achieves exceptional performance across multimedia applications, systems can fail abruptly without warning. This work shifts focus from error reduction to error prediction, enabling more reliable and trustworthy deployment.
This paper addresses the practical challenge of deploying AI agents on diverse, domain-specific workflows—from enterprise web applications requiring dozens of clicks and form fills, to multi-step research pipelines, code review across unfamiliar repositories, and customer escalations. Rather than requiring expert-driven custom harnesses for each new task domain, the authors propose methods to enable AI agents to generalize and adapt.
Paul Graham analyzes how superlinear returns operate in business and entrepreneurship, explaining why exceptional work generates disproportionately large rewards and how this principle applies to competitive advantage and startup success.
Paul Graham provides a framework for identifying and pursuing great work, covering passion discovery, skill development, and the mindset required to make meaningful contributions in your chosen field and build a remarkable career.
Paul Graham explores how to generate fresh ideas through diverse experiences, deep reading, and creating conditions for creative breakthroughs. He emphasizes the importance of curiosity, observation, and maintaining an open mind to recognize unexpected connections.
This paper explores answer set programming (ASP) and its quantifier extension ASP(Q) for querying inconsistent prioritized data in knowledge bases. It proposes three notions of optimal repairs—Pareto-optimal, globally-optimal, and completion-optimal—to handle conflicting facts systematically.