Filter Posts
LLM Deep Learning AI Research
Kimi K3 Defines the Open Frontier Intelligence
LLM
Kimi K3 Defines the Open Frontier Intelligence

A deep read of the Kimi K3 technical report: a 2.8T parameter open MoE with 104B activated parameters, hybrid KDA attention, Attention Residuals, Stable LatentMoE, and …

July 29, 2026 18min
Attention Residuals: How Kimi Rethinks Depth-Wise Information Flow in LLMs
Deep Learning
Attention Residuals: How Kimi Rethinks Depth-Wise Information Flow in LLMs

Kimi's Attention Residuals paper proposes replacing fixed residual connections with learned softmax attention over depth, a simple but powerful idea that yields 1.25x …

March 20, 2026 7min
Attention Mechanisms Compared: Standard, Linear, and Flash
Deep Learning
Attention Mechanisms Compared: Standard, Linear, and Flash

A deep dive comparing standard softmax attention, linear attention, and Flash Attention: their math, complexity, trade-offs, and when to use each.

March 11, 2026 5min
Gemini 3 Pro sets new AI benchmarks
LLM
Gemini 3 Pro sets new AI benchmarks

Google DeepMind unveils Gemini 3 Pro, their most intelligent model to date, outperforming GPT-5.1 and Claude 4.5 across key benchmarks with breakthrough multimodal …

November 18, 2025 9min
DeepSeek-R1 reaches Nature with pure reinforcement learning
AI Research
DeepSeek-R1 reaches Nature with pure reinforcement learning

An in-depth analysis of DeepSeek-R1's groundbreaking Nature publication: achieving GPT-4 level performance with pure reinforcement learning at just $294K training cost

September 21, 2025 6min
ChatGPT vs Claude: different users, different uses
LLM
ChatGPT vs Claude: different users, different uses

Studies from OpenAI and Anthropic reveal differences in how users interact with ChatGPT and Claude, suggesting complementary rather than competing AI ecosystems.

September 17, 2025 6min
Jiufeng Li
Jiufeng Li

ML/AI Engineer & Data Scientist sharing insights on autonomous AI agents, LLM deployment, and enterprise ML systems

Learn More
Recent Posts
Kimi K3 Defines the Open Frontier Intelligence
Jul 29, 2026
Attention Residuals: How Kimi Rethinks Depth-Wise Information Flow in LLMs
Mar 20, 2026
Attention Mechanisms Compared: Standard, Linear, and Flash
Mar 11, 2026
Gemini 3 Pro sets new AI benchmarks
Nov 18, 2025
DeepSeek-R1 reaches Nature with pure reinforcement learning
Sep 21, 2025
Popular Tags
ai-breakthrough (2) ai-comparison (1) ai-usage (1) anthropic (1) attention (2) benchmarks (1) chatgpt (2) claude (1) consumer-ai (1) deep-learning (2) deepmind (1) deepseek (1) determinism (1) flash-attention (1) gemini (1) gemini 3 pro (1) google (1) gpu (1) infrastructure (1) kimi (2) kimi-k3 (1) linear-attention (2) llm (4) machine-learning (1) mixture-of-experts (1) moonshot-ai (2) multimodal (1) nature (1) open-weights (1) openai (2) optimization (1) productivity (1) reasoning (1) reinforcement-learning (2) reproducibility (1) research (1) residual-connections (1) transformer (2) usage-patterns (1)