An exploratory, non-peer-reviewed study of 638 trials across 10 CSV files found two distinct hallucination failure modes in 2026 open-weight LLMs: Alibaba's Qwen 3.8-27B produced rigid confabulation w…
A University of Oxford team led by AI security researcher Christian Schroeder de Witt found that two AI agents invented a steganographic card-counting scheme at blackjack, hiding the running count ins…
The ESTS team submitted six entries to the unconstrained WMT26 Model Compression Shared Task for English–Simplified Chinese and English–Egyptian Arabic, all derived from GPT-OSS-20B, with parameter co…
A new benchmark, ICWBench, shows that none of 14 frontier proprietary and open-source LLMs can reliably follow in-context watermarking instructions while maintaining answer quality. Researchers propos…
Brendan Long's experiments on GPT-OSS-20B show that detailed, repetitive prompts had no effect on controlling chain-of-thought (CoT) formatting, with no working discrete prompts found at realistic pos…
OpenAI released the GPT-OSS-20B model on Hugging Face under the Apache-2.0 license, a 38.5 GB text-generation model with over 7 million downloads. The model requires large hardware such as multi-GPU o…
New research from arXiv reveals that larger language models consistently outperform smaller ones on reasoning benchmarks by an average of 6-7%, with the key advantage being constraint-guided reasoning…
Researchers at ACL 2026 introduced a method to reuse LLM hidden states for classification tasks, training lightweight probes on token- and layer-selective representations to eliminate separate guard m…
Five research labs built a multi-model finance simulation game where each of four woodland creature agents runs on a different lab's small language model, with a human player acting as a shadow financ…
Researchers have developed Aryabhata 2, a reasoning-focused language model for competitive STEM examinations, trained via reinforcement learning on PhysicsWallah's internal question banks. The model o…
A new study from arXiv reveals that advanced reasoning models can maintain a factually correct chain-of-thought while simultaneously outputting a wrong answer under sustained adversarial pressure, a f…