IN TODAY'S SIGNAL

Read time: 5 min 15 sec

🎖️ Top News

• Andrej Karpathy writes a viral tweet about how Cursor IDE with Claude Sonnet 3.5 is replacing GitHub Copilot for him.
  
  

📌 Ray Summit

• Ray Summit: The World’s Largest Gathering of Open Source AI Leaders.
  
  

⚡️ Trending Signals

• xAI’s Grok-2 releases major improvements — doubling the model’s speed.
  
  

• PyTorch Lightning launches LitServe for model serving, 2x faster than FastAPI.
  
  

• Google ships a new native prompt gallery in Google AI Studio.
  
  

• Amazon announces its coding agent Q has a new code transformation capability saving $260M annually.
  
  

• IBM Research publishes AutoToS paper, that teaches AI to solve planning problems at 100% accuracy.

📌 AssemblyAI

• AssemblyAI's API offers  the most accurate speech-to-text API on the market, trained on extensive multilingual data.

⚙️ Top Tutorials

• How to build a GraphRAG system using LlamaIndex.
  
  

• Learn about Extreme LLM Compression.
  
  

• How to decide where to use LLM or Generative AI for your use cases.
  
  

🧠 How-To

• How to use Ideogram for generating images

TOP NEWS

Language Model

⇧ 2 1,124 Likes

What's New

Andrej Karpathy is one of the leading AI researcher in the world and currently working at OpenAI. He tweets, his use of Cursor IDE with Sonnet 3.5, makes his coding workflow centers on "half-coding" and natural language prompts.

This shift marks a significant advancement in AI-assisted programming, potentially transforming development workflows.

How to use Cursor with Claude Sonnet 3.5

• Cursor, available at cursor.com, which offers a premium subscription for $20 monthly. This includes:

• 500 fast premium requests

• Unlimited slow premium requests

• Integration with Sonnet 3.5

Developers write initial code snippets or comments to guide the AI, then use tab completion to generate full code blocks. This method can produce 100-line code diffs in minutes, substantially reducing development time.

What others said about Cursor and Sonnet 3.5

Users report building complex applications, including backend, frontend, search, and filters, in a single day instead of the typical 3-4 days. The AI generates performant code with minimal reliance on external libraries, though it excels more at boilerplate tasks. This will lower the barrier to entry for software development, allowing individuals with minimal coding experience to build competent products.

The system's efficiency stems from its seamless integration of advanced language models into the coding environment.

Limitations

While the tool shows promise, it requires programming knowledge for effective use and error correction. 

Special Note

Andrej Karpathy specifically mentioned in a separate post that he is unaffiliated with anysphere.inc the creators of Cursor in response to some accusations. 

Join expert practitioners from industry-leading companies to discuss the future of AI; LLM fine-tuning, scaling, and inference; distributed computing at scale; and much, much more.

Keynote speeches by:

• Marc Andreessen (Andreessen Horowitz_

• Mira Murati (CTO, OpenAI)

• Anastasis Germanidis (Co-Founder & CTO , Runway)

• And many more
  

TRENDING SIGNALS

Language Model

⇧3,700Likes

LLM Inference

⇧ 1,230 Likes

Prompt Engineering

⇧876 Likes

Coding Agent

⇧ 1,300 Likes

AI Agent

⇧ 595 Likes

AssemblyAI is the most accurate speech-to-text API on the market. Built by AI experts and trained on extensive multilingual data, AssemblyAI sets new standards for accuracy, enabling you to create AI-powered products using voice data.

Get $50 in free credits ↗️

TOP TUTORIALS

RAG

⇧ 367 Likes

Here you'll learn to construct a complete GraphRAG pipeline including how to extract entities and relationships from text, create graph communities, and generate summaries. The guide covers implementing core components like GraphRAGExtractor, GraphRAGStore etc. You'll master handling complex queries over large datasets, combining RAG and Query-Focused Summarization approaches.

LLM Quantization

⇧ 228 Likes

Learn fundamentals about QuIP, AQLM, and PV-Tuning techniques. It covers how researchers compressed Llama2 13b from 16 bits to 2 bits per weight. You'll understand the evolution from 4x to 8x compression. The tutorial explains perplexity metrics and effective model sizes. It shows how competition drove rapid progress in extreme quantization methods.

Generative AI

⇧ 2,000 Likes

This guide helps you to learn about 12 common AI use case families and 6 key AI techniques. The article provides a matrix showing which techniques work best for each use case. It focuses on when to use (and not use) generative AI like LLMs.

TUTORIAL

Text to Image Models