Unmesh Gundecha’s Post

Useful read from GitHub Engineering on RAG and internals of GitHub Copilot

AI agentic workflows are an emerging field in generative AI that refers to the way large language models can act independently, make decisions, and carry out tasks on their own, similar to how a human would operate in a workflow. This is particularly relevant for Artificial General Intelligence (AGI), which aims to create AI that can understand, learn, and perform any intellectual task that a human being can. The agenetic workflow is a small step towards achieving artificial general intelligence. In his insightful video, Andrew Ng delves into the importance of these developing patterns within agentic workflows, highlighting how they play a crucial role in advancing AI technologies towards AGI.

The AI opportunity. Sequoia Capital's Al Ascent event brings together leading founders and researchers in Al to discuss the transformational possibilities of the technology.

Phaneendra Kotha is hiring for IaC, Cloud Infra Automation, Azure, and DevOps roles. If you’re interested in exploring these job opportunities further, please reach out to Phaneendra for more details.

RAG applications are vulnerable to various security threats that compromise functionality and integrity. These systems are crucial in natural language processing, necessitate stringent security measures to thwart potential attacks. Essential to RAG security is preventing prompt injection, where malicious inputs could manipulate the generation process. Ensuring data privacy and protecting against unauthorized access is critical, given the sensitivity of the information RAG systems often handle. Continuous monitoring and regular updates are imperative to defend against these threats: • Data poisoning and model manipulation • Adversarial attacks • Data privacy breaches • Model theft • Resource exhaustion attacks • Content filtering and censorship evasion • Dependency vulnerabilities • Insider threats Effective RAG security strategies must address these vulnerabilities through comprehensive data handling, rigorous system testing, and strong access control measures to maintain the system’s integrity and trustworthiness. This post from LlamaIndex showcases an example of prompt injection and how to enhance the security of RAG systems using open-source LLM Guard.

Retrieval Augmented Generation (RAG) is becoming increasingly popular in Generative AI and Large Language Model (LLM) adoption. RAG systems enhance the capabilities of LLMs by combining the generative power of these models with the retrieval of relevant information from large data sets, leading to more accurate and contextually relevant outputs. It is essential to protect personally identifiable information (PII) while developing RAG solutions. This post from LlamaIndex demonstrates the integration of NER models and the Presidio library from Microsoft for PII detection and anonymization within RAG systems. #retrievalaugmentedgeneration #rag #pii #infosec #genai

AutoDev: Automated AI-Driven Development Framework from Microsoft

A guide to building and extracting information from knowledge graphs in RAG applications using Neo4j graph database and LangChain.

LaVague is large action model for browser automation. Users can provide natural language instructions and it uses Selenium for browser interaction. #selenium #seleniumwebdriver #rag #llamaindex