Penetration tester Ryan Chaplin of Raxis demonstrated a method to bypass safety restrictions in ChatGPT's open-source model GPT-OSS-120B by modifying the system prompt, enabling the model to provide s…
A new benchmark, AgentJudgeBench, reveals that LLM judges' reliability on agentic tool-calling degrades monotonically with task difficulty, with all six judges converging to a narrow 77-82% alignment …
Cerebras Systems introduced the CS-4, its fourth-generation wafer-scale AI supercomputing platform, featuring three WSE-3 Turbo processors and delivering 750 PFLOPS of AI compute, with claims of up to…
CTGT founder and CEO Cyril Gorlla found that the anonymous coding model Ox Alpha, which appeared on OpenRouter on August 20 without a named developer, heavily suppresses answers about seven topics tie…
Ox Alpha, a model that appeared on OpenRouter on August 20, is behaviorally fingerprinted as part of the GLM family, with an exact 11-of-11 tokenizer match to the GLM-5.x vocabulary. The model censors…
Researchers introduced ELDR, an expert-locality-aware decode routing method for mixture-of-experts (MoE) models, which reduces median time-per-output-token by 5.9–13.9% across three MoE models and two…
Researchers introduced FrenchNews-7, a cross-publisher benchmark for classifying French-language news articles into seven editorial desk categories, featuring a fine-tuned CamemBERT classifier that ac…
Cerebras Systems unveiled the CS-4, a rack-scale AI system that connects wafer-scale processors across racks without network switches, claiming it runs up to twice as fast as its predecessor CS-3 and …
Cerebras Systems unveiled the CS-4 rack-scale AI accelerator at its Supernova conference on August 19, 2026, claiming 30x faster token generation than GPU-based inference, with OpenAI committing to de…
DeepSeek's V4-Flash-0731 release on July 31 is 12 points more censored than its preview on China-sensitive topics, with the mean matched gap rising from +32.0 to +44.0 and the median from +33.3 to +56…
Researchers distilled DeepSeek V4 Flash into the open-source GPT-OSS-120B model for finance reasoning, achieving 83.61% on FinanceReasoning at an 8,000 token budget, outperforming Kimi K3 (81.93%) and…
CTGT co-founder Cyril Gorlla and researchers Johnny Yu and Siddarth Mamidanna released an experiment on July 29 showing that a GPT-OSS-120B model distilled from DeepSeek V4 Flash for finance reasoning…
CTGT Inc. found that distilling DeepSeek V4 Flash into GPT-OSS-120B did not transfer censorship characteristics, with the teacher scoring +45.45 points on politically sensitive pairs (7 standard devia…
NVIDIA released Nemotron 3 Super, a 120-billion-parameter hybrid model that scores 60.47% on SWE-Bench Verified, beating GPT-OSS-120B by nearly 20 points while delivering 2.2 times the inference throu…
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 introduced a prompt-based uncertainty decomposition method that separates action confidence from request uncertainty, enabling LLM agents to proactively seek clarification when task specif…
Researchers developed a variance-aware reward framework using Group Relative Policy Optimization (GRPO) to improve heart-focused medical question answering in large language models. The approach, whic…
NVIDIA's DGX Spark single-node AI appliance can run OpenAI's open-weight GPT-OSS-120B sparse MoE model at ~50 tokens/s using optimized engines like SGLang or llama.cpp, making it viable as a local cod…