AI
Large models are powerful, but solving deep, multi-step problems is pricey. ToolOrchestra trains a small “orchestrator” that decides which tools to use, when, and how—like a smart project manager for AI. It uses reinforcement learning with rewards for outcomes, efficiency, and user preferences. The result: an 8B Orchestrator that
LLM
What’s new Adding the right metadata can make LLM pretraining faster and more effective—and it’s not just about URLs. * Fine‑grained signals work: Prepending detailed quality indicators helps models learn quicker. * Append-and-predict: Appending metadata and training the model to predict it as an auxiliary task boosts efficiency.
AI
Small model, big wins Large language models are great generalists, but really tough, multi-step problems still strain both brains and budgets. ToolOrchestra flips the script: instead of one giant model, a small “orchestrator” coordinates other models and specialized tools. Trained with reinforcement learning that rewards outcomes, efficiency, and user preferences,
AI
What’s new As AI "teams" of LLM agents grow, the cloud bill can explode. BAMAS is a framework that designs these teams with a dollar cap in mind. How it works * Picks the right mix of models: Uses integer linear programming to balance accuracy vs. price. * Plans
Cybersecurity
Can AI spot phishing? This new email dataset puts it to the test Phishing and spam are getting smarter—often written by large language models. Rebeka Toth, Tamas Bisztray, and Richard Dubniczky release a labeled email dataset that separates phishing, spam, and legitimate messages, and flags whether they were written
Cybersecurity
Smarter email defenses, grounded in real messages Phishing and spam are evolving fast—often with help from AI. This study releases a large, carefully labeled email dataset spanning phishing, spam, and legitimate messages, with a key twist: it marks whether each message was written by a human or an LLM.
AI
How do we know if an AI model is truly aligned with human preferences—helpful, honest, safe, and instruction-following? This paper explores a surprisingly effective shortcut: judge the judges. Instead of grading a model’s open-ended answers (which needs lots of human effort or very strong AI judges), the authors
AI
Which AI better helps pharmacists-in-training? Researchers tested DeepSeek-R1 and ChatGPT-4o on 2,306 real, text-only questions (2017–2021) from China’s pharmacist licensure exam, scoring exact-answer accuracy in Chinese. * DeepSeek-R1: 90.0% accuracy vs. ChatGPT-4o: 76.1% (p<0.001) * Advantages held across units, especially foundational and clinical synthesis
AI
Can AI think creatively? Large language models are great at step-by-step logic, but real-world breakthroughs need more than routine problem-solving. This paper proposes a creative reasoning framework, rooted in cognitive science, and turns it into practice with "Universe of Thoughts" (UoT): methods that help LLMs search wider and
AI
How do we know an AI is truly helpful, honest, safe—and follows instructions? Today, we usually read its answers and score them, which can be slow, costly, and subjective. This paper finds a simple clue: when compared to a trusted reference, models that write well also judge well. The
AI
LLMs can write tidy code that passes unit tests—but can they build money-making agents that plan, bid, and deliver under pressure? This study pitted 40 LLM-coded agents (prompted with methods including “vibe coding,” i.e., high-level guidance) against 17 human-coded agents from grad CS students in a real-world-style challenge:
AI
TL;DR An open-source platform helps creators check if their content was used to train large language models, without heavy compute or opaque workflows. What it does * Verifies if a specific work appears in LLM training datasets. * Improves similarity matching and dataset validation. * Cuts compute by 10-30% via efficient API