Patryk Kaczmarek

We made a few changes to our "RAG system from the scratch" v1. The most notable changes being - 1. Updated TF-IDF matrix logic for embeddings to context-preserved sentence/document embeddings using intfloat/e5-small. 2. Updated Summarizer with a new model — facebook/bart-large-cnn, also increased the max length, improving the context preserved after summarization. A lot happened before we published v2 to Github and its all documented for a good read / laugh / learning. What ever amuses you today, pick that tone! Check the link in the comments. #AI #Startup #GenAI

1

1 komentarz

𝐂𝐡𝐞𝐚𝐭 𝐒𝐡𝐞𝐞𝐭: 𝐀/𝐁 𝐓𝐞𝐬𝐭𝐢𝐧𝐠 𝐟𝐨𝐫 𝐀𝐈 𝐌𝐨𝐝𝐞𝐥𝐬 A/B testing in AI models helps improve performance by comparing two model versions to understand what works best. Here’s a simplified guide 👇 🗞️ Get Free Daily Updates on the Latest AI News -> https://lnkd.in/dH8ag7MA #ABTesting #MachineLearning #AIModels #DataScience

575

24 komentarze

Curious about the speed and performance of Rust vs. C++? Dive into our benchmarks and analysis to see which language comes out on top. Read the full article for a detailed comparison: https://lnkd.in/djqiQUqi #Rust #CPlusPlus #Programming #TechTrends #Coding

3

How to supercharge RAG applications with graph-based metadata filtering? Want to take your Retrieval-Augmented Generation (RAG) applications to the next level? This insightful blog post by Tomaz Bratanic from Neo4j explores how graph-based metadata filtering can significantly improve vector search accuracy and unlock the full potential of your knowledge graphs. 🤯 Challenges of traditional vector search: While text embeddings and vector similarity search are powerful tools for finding relevant documents, they often struggle with structured criteria like dates, categories, or specific relationships between entities. This is where metadata filtering comes in! Benefits of graph-based metadata filtering: - Enhanced accuracy: Structured filters can be used to narrow down the search space before applying vector similarity search, leading to more precise and relevant results. - Unlock hidden relationships: Leverage the power of graph databases to explore complex connections between entities and refine your search based on specific criteria. - Dynamic query generation: Use LLMs to generate Cypher statements on the fly, adapting to user input and retrieving the most relevant information from your knowledge graph. The blog post provides a practical example using LangChain and Neo4j, demonstrating how to: 1)Build a custom tool that dynamically generates Cypher statements based on user input. 2)Implement metadata filters for organization, country, sentiment, and topic. 3)Integrate the tool into an OpenAI agent for a conversational RAG experience. This approach opens up exciting possibilities for building more powerful and efficient RAG applications across various domains. Read the full blog post: https://lnkd.in/gYUj4WEF _____________ ✔️ Click "Follow" on the Cohorte page for daily AI engineering news.

3

1 komentarz

C++Now 2024- Jonathan Müller: "A Deep Dive Into C++ Object Lifetimes" youtu.be/oZyhq4D-QL4 A C++ program manipulate objects, but it is undefined behavior if you attempt to manipulate them while they are not alive. So let's do a deep dive into object lifetime. When are objects created and when are they destroyed? How does temporary lifetime extension come into play and what changed there recently? What happens when you std::malloc memory and just pretend objects are there without creating anything? Or worse: You use mmap() to read shared memory. How do unions interact with constructors, strict aliasing, or the "common initial sequence"? What when you explicitly call the destructor and later re-use the same storage? What's the deal with std::launder, std::bit_cast, and std::start_lifetime_as? We'll answer all of those questions and much more. We'll do that by looking at the C++ standard, old and new proposals, and compiler optimizations. --- Jonathan Müller is a library developer at think-cell. In his spare time, he works on various C++ open source libraries for memory allocation, cache-friendly containers, or parsing. He also blogs at foonathan.net and is a member of the C++ standardization committee.

35

Supercharge Your RAG Applications with Graph-Based Metadata Filtering! 🧠 Want to take your Retrieval-Augmented Generation (RAG) applications to the next level? This insightful blog post by Tomaz Bratanic from Neo4j explores how graph-based metadata filtering can significantly improve vector search accuracy and unlock the full potential of your knowledge graphs. 🤯 Challenges of traditional vector search: -While text embeddings and vector similarity search are powerful tools for finding relevant documents, they often struggle with structured criteria like dates, categories, or specific relationships between entities. This is where metadata filtering comes in! Benefits of graph-based metadata filtering: - Enhanced accuracy: Narrow down the search space using structured filters before applying vector similarity search, leading to more precise and relevant results. - Unlock hidden relationships: Leverage the power of graph databases to explore complex connections between entities and refine your search based on specific criteria. - Dynamic query generation: Use LLMs to generate Cypher statements on the fly, adapting to user input and retrieving the most relevant information from your knowledge graph. The blog post provides a practical example using LangChain and Neo4j, demonstrating how to: Build a custom tool that dynamically generates Cypher statements based on user input. Implement metadata filters for organization, country, sentiment, and topic. Integrate the tool into an OpenAI agent for a conversational RAG experience. This approach opens up exciting possibilities for building more powerful and efficient RAG applications across various domains. Read the full blog post: https://lnkd.in/gYUj4WEF #RAG #MetadataFiltering #GraphDatabases #LangChain #Neo4j #AI ✔️ Click "Follow" on the Cohorte page for daily AI engineering news.

6

Rust vs. C++: Which Language Should You Choose? 🚀 When building your next project, should you opt for Rust’s modern safety features or C’s reputable low-level control? 🤔 Check out our latest guide to see how these two programming titans stack up across speed, memory safety, and versatility! https://lnkd.in/gcrPc_3f #RustLang #CProgramming #Developers #WebHosting

1

2 komentarze

✨Build a Knowledge Graph-based Agent With Llama 3.1, NVIDIA NIM, and LangChain Great blog by Tomaz Bratanic! Use Llama 3.1 native function-calling capabilities to retrieve structured data from a knowledge graph to power your RAG applications https://lnkd.in/gnW3RXXK

511

4 komentarze

Vector databases are so hot right now. Even databases that weren't originally made for vectors are getting vector capabilities added via plugins. This blog goes over Dolt's plan for Vectors. https://lnkd.in/dhc3cfZ3

7

1 komentarz

Hey #Ronin devs! Archive requests support with Debug & Trace are now available on Chainstack Global Elastic Nodes 💙🛠️ Here’s how this improves your workflow: 🌍 Lightning-fast access to all historical blockchain states. 🌎 Monitor, audit, and replay on-chain activity effortlessly. 🌏 Global fault-tolerant, geo-load-balanced infrastructure. Full story ⤵️ https://lnkd.in/dg9QGSTM

1

A Survey of RAG and RAU: Advancing Natural Language Processing with Retrieval-Augmented Language Models Quick read: https://lnkd.in/gnGqk44W Paper: https://lnkd.in/g8-VftbX #llms #largelanguagemodels #ai

20

Check Out This *Easy-To-Understand No-Code* Automated Trading Strategy for NinjaTrader ®: Bollinger Bands ® Long Entry on Breakout Above Upper Band (With X-Axis Offset). No NinjaScript ® programming knowledge required. https://lnkd.in/guGyjHQx #NinjaTrader #NinjaScript

1

💻 Eager to learn the internals of ARM64 architecture, reverse engineering, and binary exploitation? Read our comprehensive blog series at ➡️ https://lnkd.in/drrBm_UH Part 1 - ARM Instruction Set + Simple Heap Overflow Part 2 - Explore Use-after-free Vulnerability Part 3 - A Simple ROP Chain Part 4 - Using Mprotect() to bypass NX protection Part 5 - Writting Shellcode Part 6 - Exploiting an Uninitialized Stack Variable Vulnerability Part 7 - Bypassing ASLR and NX Part 8 - Exploiting an Integer Overflow Vulnerability Part 9 - Exploiting an Off by One Overflow Vulnerability Part 10 - Intro to Arm Memory Tagging Exploitation (MTE) Start learning at https://lnkd.in/drrBm_UH #ARM64 #Reversing #Exploitation #RISC #ReturnOrientedProgramming #Shellcodes #iosSecurity #AndroidSecurity #MobileSecurity #CyberSecurity

60

2 komentarze

Did you know the OPTX file used in Quadoa Optical CAD is coded using the text readable #xml format? It readily works with the same #git version control that nearly all software developers use today. This allows multiple user's design changes to be tracked, differences to be resolved, and historical versions to be restored as desired. There is no need to save renamed files or use other antiquated version control methods. Find out about more features that make #Quadoa next-gen optical design software. #optics #opticaldesign www.dyoptr.com/products

6

Neo4j is the world’s most popular #GraphDatabase for good reasons. Robust, reliable and efficient, it's the perfect partner for ReGraph, our #GraphVisualization SDK for React. https://hubs.li/Q02CMSMR0 This three-step tutorial, authored by graph expert, Kavita, shows you how to build a scalable graph application that's designed for your data, your users and the questions they need to answer.

1

Meta Unveils Llama 3: Advancing Open Source Large Language Models Key Points - Meta introduces Llama 3, the latest iteration of its open source large language model (LLM), claiming it surpasses previous benchmarks, including ChatGPT-3.5. - Initial models released are 8 billion and 70 billion parameters, with larger models over 400 billion parameters planned for future release. - Llama 3's development spanned over two years, focusing on high-quality training data, distributed training scalability, model architecture optimization, and instruction fine-tuning. - Meta's 70 billion parameter model outperformed competitors in human evaluations across key usage scenarios like coding, reasoning, and creative writing. - The company adopts an "open by default" approach, making Llama 3 available across various platforms, cloud providers, and model hosts. - Accompanying the release are updated AI safety tools, including Llama Guard for risk classification, CyberSec Eval for assessing misuse, and Code Shield for filtering insecure code suggestions. - Meta emphasizes responsible AI development, highlighting the need for a holistic approach to address challenges like data privacy and algorithmic bias. - The end-user product integrating Llama 3 is Meta AI, claimed to be the world's leading AI assistant, available across Meta's platforms and the web. - Multimodal versions of Meta AI with vision capabilities are in development, with an early preview for Ray-Ban smart glasses. - While some express skepticism over Meta's motivations, Llama 3's release sets a new standard for openly-available LLMs, following the benchmark set by competitors like Mistral AI's Mixtral 8x22B. Llama 3 signifies a significant leap towards open and accessible AI. Ready to harness the power of AI for your business? Contact us today to learn more about our AI development services and explore how cutting-edge LLMs can transform your projects: https://lnkd.in/dFxwXDFq Read more, here: https://lnkd.in/gk-kJGbJ Additional Resources - Meta: https://about.meta.com/ - ChatGPT-3.5: https://chat.openai.com/ - Mistral AI: https://mistral.ai/ - Llama 3: https://lnkd.in/e435P2mQ Image Credits: Artificial Intelligence News #Meta #Llama3 #OpenSourceLLM #LargeLanguageModels #AI #ArtificialIntelligence #MachineLearning #GPT3 #Benchmarking #AIdevelopment #DistributedTraining #ModelArchitecture #InstructionFineTuning #AIsafety #AIethics #MistralAI #DataPrivacy #AlgorithmicBias #MetaAI #MultimodalAI #RayBan #AIdevelopmentServices #unitedstates #unitedkingdom #netherlands #uae #canada #markovate 

9

🚀 MongoDB + C++: Unleashing NLP Potential in Modern Apps What makes chatbots, virtual assistants, and sentiment analysis tools so intelligent? In our newest piece, we break down the fundamentals of #NLP—from tokenization and part-of-speech tagging to the advanced Transformers and #BERT models. These innovations are the backbone of systems that help us translate languages, understand customer sentiment, and summarize text effortlessly. Our latest article unpacks: • Core NLP Concepts: Tokenization, Part-of-Speech Tagging, Named Entity Recognition, and more. • Key Techniques: From traditional statistical models to advanced deep learning frameworks like Transformers and BERT. • Challenges and Solutions: Managing ambiguity, data quality, and addressing bias. Read the full article to see how #MongoDB and #C++ are optimizing these NLP techniques for modern applications. ☞ https://lnkd.in/diDszXHz #AI #MachineLearning #Innovation #TechTrends #DataScience #TechInnovation #FutureOfAI #Elinext #BlogArticle #ITOutsourcing

45