You understand the difference between a good IIT-JEE coaching teacher and a good university professor only much later in life. Both teach you to solve problems, but they prepare you for two very different battles.
A good coaching teacher makes you sharper. Faster. More precise. They teach you to master techniques and attack difficult problems with confidence.
A good university professor challenges your method of thinking, builds your foundations, and teaches you that asking a good question can be as important as solving a difficult problem. A good professor also leaves room for doubt. Because in science, nothing should be accepted because someone has given you an answer.
Bihar once had an extraordinary intellectual tradition in mathematics, physics and chemistry. Prof. H. C. Verma taught at Science College, Patna. I was fortunate to study under some of the professors of that generation like Asit Das Gupta, Deo Mukherjee and K. C. Sinha.
People respect you more when they don't see you often. Even parents. Trust me. It's strange how distance rearranges love, how absence restores what closeness erodes. When people are deprived of your presence, they start seeing you clearly again, not through habit but through awareness. Proximity dulls perception. Space sharpens it. That's just how the human mind works.
In a parallel universe, Yu Deng didn’t become a mathematician but instead a professional Go player. Quiet, ambitious, and focused, it was the nearest of misses that set Deng on his current path to the Fields Medal. https://t.co/Xkgpp5M29M
My profile of new Fields Medalist Hong Wang, who proved the 3D Kakeya set conjecture last year, among other blockbuster results. Interesting person and math. https://t.co/rLHjWZ6Lmf
月之暗面 CEO 揭秘下一代 AI 模型竞争逻辑
AI 竞争正在进入新阶段
不再只是比参数规模
而是比谁能真正解决复杂任务
月之暗面 CEO 提出三个关键方向:
Token效率提升
用更少的计算成本,完成更高质量推理
无损长文本能力
让 AI 能理解更长上下文,不只是“记住”,而是真正调用信息
Agent智能体集群
多个 AI 协同工作,完成过去需要人工拆解的复杂流程
未来的大模型竞争
拼的不是谁的参数更多
而是谁能在真实世界里
把任务完成得更快、更准、更可靠
#Moonshot #Kimi #AI
GOOGLE OPEN-SOURCED A LIBRARY THAT TURNS MESSY TEXT INTO STRUCTURED DATA
langextract uses LLMs to pull structured info out of unstructured documents like clinical notes and reports, and it doesn't just guess, it maps every extraction back to its exact location in the source text
→ precise source grounding, so every extracted entity links to where it appeared in the original text
→ enforces a consistent output schema from your few-shot examples, using controlled generation on models like Gemini
→ built for long documents, using chunking, parallel processing, and multiple passes to avoid missing things buried deep in the text
→ generates a self-contained interactive HTML visualization to review thousands of extractions in context
→ works with Gemini, OpenAI models, and local models through Ollama, no model fine-tuning required
pip install langextract and you're extracting in a few lines of code
https://t.co/nUt8IXL9d5