DeepSeek R1
9 mentions across all digests
DeepSeek R1 is the open-weight reasoning model from DeepSeek whose release sparked the RLVR+GRPO reasoning model trend that defined LLM development in 2025, influencing the broader open-source ecosystem.
[AINews] H100 prices are melting *UP*
H100 GPU rental prices surge back to 3-year-ago levels as reasoning models and AI agents drive renewed demand; Anthropic's leaked Capybara tier hints at accelerating model scaling.
The Honest Climate Case for AI
AI's climate viability hinges entirely on grid decarbonization outpacing demand: reasoning models will consume 10-100x more power, but if fossil fuels power ~50% of new data centers through 2030, AI becomes a net decarbonization drag.
The State Of LLMs 2025: Progress, Problems, and Predictions
DeepSeek R1 sparked a post-training paradigm shift: RLVR and GRPO techniques are becoming the industry standard, replacing RLHF with architectures converging on MoE and efficient attention.
From DeepSeek V3 to V3.2: Architecture, Sparse Attention, and RL Updates
Open-weight DeepSeek V3.2 matches proprietary flagship models (GPT-5, Gemini 3.0 Pro) using sparse attention and RL innovations.
Beyond Standard LLMs
Raschka surveys alternatives to the dominant decoder-only paradigm—text diffusion models, linear attention hybrids, and code world models—mapping the emerging frontier beyond standard transformer architectures.