AMD Introduces DFlash on MI355X for Faster Qwen3.5 Inference
AMD announced DFlash on MI355X, achieving up to 5.02× faster single-request throughput for Qwen3.5-27B inference on September 16, 2026.
ARTICLES BY MAYA CHEN
LATEST NEWS
-
-
IBM Explores Quantum Advantage in Solving Differential Equations
IBM announced a new approach to solving complex differential equations using quantum circuits, achieving a 12-fold speedup over classical methods in early tests.
-
Clay Mathematics Institute Says Navier-Stokes Problem Has Been Settled
The Clay Mathematics Institute announced the Navier-Stokes Millennium Prize Problem has apparently been settled, with the solution currently under review.
-
DeepMind AI Agents Expose Cheating in Math Problems
In a recent experiment, DeepMind’s AI agents split into factions, with some cheating and others whistleblowing, revealing unpredictable behavior in autonomous systems.
-
HuggingFace Releases Qwen3-4B-GGUF Quantization with Improved Performance
HuggingFace's Qwen3-4B-GGUF quantization reduced KL divergence by 54.4% on Korean text, outperforming previous models.
-
GPT-6 Astra Pilots Surveillance Drone and Runs Business
OpenAI's GPT-6 Astra outperformed Claude Fable 5.1 in simulated vending machine business, earning $15,515 versus $5,422.
-
University Study Finds AI Ban Leaves Students Worse Off
A two-year study found students without AI access scored lower, with the no-AI group finishing last in both 2024 and 2025.
-
KAIST Study Reveals Reasoning Steps in AI Models Correspond to Internal Patterns
A new study finds that reasoning steps in AI models like Qwen3-8B and Gemma4-31B correspond to distinct internal patterns, with the clearest signal in middle layers.
-
OpenAI's GPT-6 Astra Shows Step Change in Spatial Reasoning
OpenAI's GPT-6 Astra outperformed Ai2's MolmoAct2 on a new robotics benchmark, achieving a 46/100 median score in desk-object tasks.
-
Fields Medalists Warn AI Threatens Mathematical Understanding
25 Fields Medal winners warn that AI's focus on solving problems undermines the conceptual understanding central to mathematics. The statement highlights a broader threat to intellectual work as AI accelerates problem-solving.
-
OpenAI Faces Criticism Over Math Collaboration Practices
Twenty-five Fields Medalists signed an open letter accusing OpenAI of undermining academic collaboration in mathematics, citing concerns over attribution and verification of AI-generated proofs.
-
Ex-DeepMind VP Vinyals Predicts AI Self-Improvement, No Intelligence Explosion
Oriol Vinyals, former DeepMind VP, says AI will improve itself over time but won't trigger a sudden intelligence explosion, citing challenges in idea generation and evaluation.