- unwind ai
- Posts
- Turn Any LLM into AI Agent
Turn Any LLM into AI Agent
PLUS: Customize Audio Overviews in NotebookLM, Opensource Apple Intelligence Writing Tools
Shubham Saboo & Gargi Gupta
October 18, 2024
Today’s top AI Highlights:
Build, ship, and monitor Agents with blazing-fast memory, knowledge, tools & reasoning
NVIDIA Nemotron beats GPT-4o, Claude 3.5 Sonnet, and Llama 405B in chat capabilities
Perplexity lets you make custom AI assistants for AI-powered research and collaboration
NotebookLM now lets you customize the Audio Overview AI podcast
Free opensource version of Apple Intelligence Writing Tools for Windows
& so much more!
Read time: 3 mins
AI Tutorials
Building a RAG app that interacts with YouTube videos might sound complicated—especially since most LLMs can’t natively process videos. But with the right tools, it’s a cakewalk.
In this tutorial, we’ll walk you through building an LLM app with RAG to interact with YouTube videos using the Embedchain framework and GPT-4o. And the best part? You can get this up and running in just 30 lines of Python code!
We share hands-on tutorials like this 2-3 times a week, designed to help you stay ahead in the world of AI. If you're serious about levelling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.
🎁 Bonus worth $50 💵
Latest Developments
Stop juggling with tools and frameworks to build smart AI agents. Phidata’s latest version makes it faster and easier to build, manage, and monitor AI agents. The update supercharges agents with faster memory and knowledge, and brings 100+ tools, reasoning, and multi-agent collaboration. A new dedicated UI now gives you full control over agents, with all data stored locally for privacy. Whether running agents on your machine or deploying to the cloud, Phidata simplifies infrastructure management with templates for fast deployment and built-in monitoring.
Key Highlights:
70% Faster Memory and Knowledge Retrieval - Agents now store and recall data quicker, boosting response speed. This is handled through optimized storage in SQLite or PostgreSQL for consistent, reliable performance across sessions.
Over 100 Tools - Integrate your agents with tools like DuckDuckGo for search and YFinance for financial data. The new version makes it easier to incorporate multiple Python functions or pre-built toolkits.
Multi-Agent Collaboration for Complex Tasks - Agents can now work together, sharing data and coordinating in real-time. This allows for tasks to be broken down efficiently across different agents, each leveraging their specialized tools and knowledge.
Interactive Playground - The new Agent UI lets you interact with agents through a ChatGPT-like interface. All data stays local for security and transparency. You can test, monitor, and debug agents directly from the playground.
Reasoning for Step-by-Step Problem Solving - The new reasoning feature, available with OpenAI o1, enables agents to think through problems logically before responding. While still experimental, this feature can help with more thoughtful responses but may not work reliably for all tasks yet.
Writer RAG tool: build production-ready RAG apps in minutes
RAG in just a few lines of code? We’ve launched a predefined RAG tool on our developer platform, making it easy to bring your data into a Knowledge Graph and interact with it with AI. With a single API call, writer LLMs will intelligently call the RAG tool to chat with your data.
Integrated into Writer’s full-stack platform, it eliminates the need for complex vendor RAG setups, making it quick to build scalable, highly accurate AI workflows just by passing a graph ID of your data as a parameter to your RAG tool.
NVIDIA quietly dropped Llama-3.1-Nemotron-70B-Instruct trained with REINFORCE for better response quality. Built on the Llama-3.1-70B base, it has been optimized to improve the helpfulness and accuracy of outputs in general-domain tasks. This model has achieved top performance across multiple benchmarks, including Arena Hard and MT-Bench, making it one of the most competitive models available today. You can integrate it easily using Hugging Face’s Transformers library.
Key Highlights:
Performance Edge - Llama-3.1-Nemotron-70B-Instruct leads the pack with Arena Hard, AlpacaEval, and MT-Bench, outperforming GPT-4o and Claude 3.5 Sonnet on all alignment benchmarks.
Optimized for Helpfulness - The model is trained with over 21,000 human-labeled prompt-response pairs, making it highly responsive to user queries. It excels in both factual correctness and coherence.
Seamless Integration - Available through Hugging Face and via API, the model can run on two 80GB GPUs (NVIDIA Ampere or newer). The model supports up to 128k input tokens and outputs responses with a length of up to 4k tokens.
Quick Bites
Perplexity has released Internal Knowledge Search for Pro and Enterprise users to search both web content and internal knowledge bases in one place with multi-step reasoning and code execution.
They have also launched Perplexity Spaces, an AI-powered research and collaboration hub where you can invite others, upload additional files as persistent sources, pick an AI model of your choice, write custom instructions, and set the A assistant to respond in the way you want. This feature is also available to Pro and Enterprise users.
This was probably the most requested feature in Google’s NotebookLM. You can now provide instructions before you generate a Deep Dive Audio Overview. For example, you can focus on specific topics or adjust the expertise level to suit your audience. Just hit "Customize" to provide instructions for the AI hosts before generating.
OpenAI has released an early version of the ChatGPT desktop app for Windows. It is available for testing for Plus, Enterprise, Team, and Edu users. OpenAI will bring the full experience to all users later this year. Get instant ChatGPT help in any app or website with the Alt + Space keyboard shortcut.
OpenAI's Chat Completions API now supports both text and audio. Pass text or audio inputs, then receive responses in text, audio, or both. You can use it to create asynchronous audio experiences or switch to the Realtime API for low-latency interactions.
Tools of the Trade
Writing Tools: A free, opensource grammar assistant for Windows, inspired by Apple Intelligence Writing Tools, that uses Gemini 1.5 Flash for advanced grammar, spelling checks, rewriting, and more. It works system-wide, is privacy-focused, customizable, and supports multiple languages.
AnotherWrapper: All-in-one Next.js AI starter kit with pre-built UI components and integrations like OpenAI, Supabase, Replicate, etc. Quickly launch AI products by bundling essential features like authentication, payments, analytics, and SEO, saving time on setup and infrastructure.
PostBot 3000: Opensource project that shows how to build a powerful AI agent, stream responses, and generate artifacts. This project makes it easier for anyone looking to implement similar solutions. Built using LangGraph, Python for AI workflows and FastAPI for creating a robust API.
Awesome LLM Apps: Build awesome LLM apps using RAG to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos through simple text. These apps will let you retrieve information, engage in chat, and extract insights directly from content on these platforms.
Hot Takes
Mira, Ilya, Elon, Sam, and Dario are now all competing with each other for AGI despite all having worked together at OpenAI just a few years ago ~
James CampbellX, where you are free to say absolutely anything — unless it’s critical of GenAI.
And then machine learning researchers will try endlessly to deplatform you (usually via ridicule rather than argument). ~
Gary Marcus
Meme of the Day
That’s all for today! See you tomorrow with more such AI-filled content.
🎁 Bonus worth $50 💵
Share this newsletter on your social channels and tag Unwind AI (X, LinkedIn, Threads, Facebook) to get AI resource pack worth $50 for FREE. Valid for a limited time only!
PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉
Reply