[태그:] Generative AI
-
DeepSeek Open-Sources DeepInfer, Its Production Inference Engine
DeepSeek open-sourced DeepInfer, the high-performance inference framework behind its production AI services, under Apache 2.0.
-
MiniMax M3.1 Flash Cuts AI Costs 70% While Boosting Speed
MiniMax launches M3.1 Flash, a fast, low-cost AI model for coding, agentic tasks, and computer use, cutting prices 70% versus its predecessor.
-
Hybrid Reasoning Agents: LLMs That Think in Latent Space
Hybrid Reasoning Agents propose LLM agents that reason internally in latent space while still acting through text, blending speed and interpretability.
-
DeepSeek-V4.1-Flash-1017 Brings Faster, Cheaper AI Reasoning
DeepSeek-V4.1-Flash-1017 is a minor but meaningful update improving reasoning, coding, and agentic tasks while cutting cost and latency.
-
FLUX 3 Image Klein: A Compact Open Weights Model for Image AI
Black Forest Labs launches FLUX 3 Image Klein, a 5.6B parameter open weights model for text-to-image, editing, and multi-reference generation on consumer GPUs.
-
Neural Networks May Be More Interpretable Than We Thought
New interpretability research argues superposition wasn’t the real cause of confusing neurons, suggesting large neural networks are easier to understand than feared.
-
TypeSafe Ships Jev 1.14.0 With Experimental “Danger Zone” Features
TypeSafe’s Jev 1.14.0 adds multimodal images, a new bounding-box type, and longer context via opt-in “danger zone” API flags.
-
Google’s Gemini 4 Argon Targets Long-Horizon Coding and Science
Google DeepMind’s Gemini 4 Argon model tackles long-horizon software engineering, cyber defense, and scientific research, from Chrome refactors to drug discovery tools.
-
Sakana AI’s MASS Lets Models Teach Themselves via Multi-Agent Teams
Sakana AI’s MASS framework lets an AI model optimize, evaluate, and train itself using multi-agent teamwork, a step toward recursive self-improvement.
-
Build Your Own Decision Model: Single-Pass LLM Inference
A developer’s guide to building constrained “decision models” that answer multiple-choice questions in one LLM pass, plus how to calibrate their confidence.