AI Research

Your Model’s MSE Is Lying to You

First in a series on probabilistic forecasting for physical signals. Next: what happens when you roll the forecast forward more than one step. The post Your Model’s MSE Is Lying to You appeared first on Towards Data…

When to Use One Model and When to Use a Team of Agents

When Codex is the right shape for the problem, when Claude Code is, and how I split 5 specialist agents between them on dense AI capacity work. The post When to Use One Model and When to Use a Team of Agents appeared…

U.S. AI Leaders Want to Slow Down. China Offers No Such Promise.

Beijing has offered no promise to match an American AI slowdown. Its response exposes the challenge facing safety advocates: reducing risk while competing with a country that rejects the proposed constraints.

AI Is Turning Intelligence Into Infrastructure. Who Owns the Pipes?

AI models are becoming cheaper and more widely available. The bigger economic question is what remains scarce—and who captures the value when model access itself becomes less exclusive.

Graph Engineering for AI Agents: From Prompts and Loops to Workflows

A viral debate over loops versus graphs points to a bigger shift in how we build AI systems. Here’s what graph engineering actually means, how it differs from prompt, context, and loop engineering, and why it matters.…

From Static to Dynamic Skills: A Different Model for Agent Knowledge

Why the skill-inflation panic is aimed at the wrong thing, and what it costs to make agent knowledge a build artifact instead of a file. The post From Static to Dynamic Skills: A Different Model for Agent Knowledge…

AI Safety’s New Paradox: America May Need a Bigger Lead to Slow Down

Trump and Anthropic’s Dario Amodei both want America to maintain its AI lead over China. What they disagree on is what that advantage is for. One wants to keep pressing it; the other sees a large enough lead as the…

AI's biggest rivals agree: slow down

OpenAI’s IPO Can Wait. Can Its Investors?

OpenAI wants the freedom to slow AI development when safety demands it. Its IPO delay raises a question for future shareholders: how much waiting will they accept?

Can AI Companies Afford to Slow Their Own Progress?

Dario Amodei wants outside evaluators inside AI labs and shared rules to slow frontier development. The test is whether companies will accept delays when rivals could gain ground.

I Am No One: Style-Aware Paraphrasing for Text Anonymization

arXiv:2609.12341v1 Announce Type: new Abstract: Authorship attribution models can re-identify users from seemingly anonymized text by exploiting stable stylistic fingerprints, even after explicit identifiers are…

PACIFIC: Can LLMs Discern the Psychometric Traits Influencing Your Preferences? Personality-Driven Preference Alignment in LLMs

arXiv:2602.07181v4 Announce Type: replace Abstract: User preferences are increasingly used to personalize Large Language Model (LLM) responses, yet reliably leveraging preference signals remains under-explored. In…

Beyond ID Embeddings: Process-Grounded Language Modeling for Cognitive Diagnosis

arXiv:2609.12403v1 Announce Type: cross Abstract: Cognitive Diagnosis Models (CDMs) play a pivotal role in personalized online learning. Traditional CDMs rely on discrete, ID-based embeddings to represent students,…

CueMem: Cue-Guided Context Reconstruction for Long-Term Conversational Memory

arXiv:2609.12354v1 Announce Type: new Abstract: Long-term conversational agents must answer user queries by recalling information from extended dialogue histories, yet directly using the full history is costly and often…

ORQA: An Occupation-Realistic Question and Answer Framework for LLM Professional Knowledge

arXiv:2609.12366v1 Announce Type: new Abstract: We present ORQA, a method for testing occupation-level knowledge in large language models. Prior methods either map abstract LLM skills to occupations via task definitions…

From Bench-to-Bedside: A Review of Clinical Trials in Drug Discovery and Development

arXiv:2412.09378v4 Announce Type: replace-cross Abstract: Clinical trials bridge basic research and clinical application, serving as essential steps in drug development. This review examines clinical trial phases (Phase…

MedSNIP: Building and Benchmarking Snippet-Level Granularity for Medical Fact Verification

arXiv:2609.12884v1 Announce Type: new Abstract: A medical claim's correctness often depends not on the claim alone, but on the clinical structure around it. A claim may require a lab reference range, a causal or…

HypoKG: Evidence-Disciplined Biomedical Hypothesis Generation Beyond Endpoint Knowledge

arXiv:2609.12260v1 Announce Type: new Abstract: Large language models (LLMs) can generate biomedical hypotheses, but it remains unclear whether they truly reason from scientific evidence or simply produce…

Continue, Adapt, or Yield: In-Turn Adaptation to Overlapping Speech in Full-Duplex Agents

arXiv:2609.13117v1 Announce Type: new Abstract: Full-duplex evaluation often emphasizes whether an agent keeps speaking or stops. That binary cannot express a third response humans use routinely: continuing to speak…

Measuring Pragmatic Influence in Large Language Model Instructions

arXiv:2602.21223v2 Announce Type: replace Abstract: It is not only what we ask large language models (LLMs) to do that matters, but also how we ask them. Phrases like ``This is urgent'' or ``As your supervisor'' can…