Today’s top AI Highlights:

1. Hermes 3 models - Llama 3.1 unleashed; less censorship and better capabilities

2. Build knowledge-grounded genAI apps with this opensource toolkit

3. Want to build the next TikTok? Just steal everything, says ex-Google CEO Eric Schmidt

4. AI code review agent in VS Code that helps you write error-free code

& so much more!

Read time: 3 mins

Latest Developments

Fine-Tuned Llama 3.1 Aligned to You ⚙️

Nous Research, known for its Hermes series of fine-tuned Llama models, has just released Hermes 3. These new models, available in 8B, 70B, and 405B parameter sizes, are built on Llama 3.1 and offer significant improvements in roleplaying, agentic tasks, and multi-turn conversations. Hermes 3 models are designed with a strong emphasis on user alignment and steerability, allowing for greater control and customization compared to commercially available models. The enhanced function calling, structured output capabilities, and improved code generation are particularly useful.

Key Highlights:

1. Performance and User Alignment - Hermes 3 models perform competitively with Llama-3.1 Instruct models, and prioritize user alignment over strict company policies. This results in less censorship and greater control for the end user.

2. Agentic Capabilities and Roleplaying - Hermes 3 features advanced agentic capabilities, including XML handling and scratchpads, along with significant improvements in roleplaying and internal monologue abilities.

3. Improved Reasoning and Multi-Turn Chat - The models showcase enhanced reasoning capabilities, step-level planning, and improved long-context coherence, a noticeable improvement over both previous Hermes versions and Llama 3.1.

4. Opensource - The weights of all Hermes 3 models are available on Hugging Face. The code for the Hermes Pro Large Language Model for function calling is also available.

Opensource Toolkit to Build Knowledge Graphs 🕸️

Stop battling LLM hallucinations and context limitations! Neo4j’s new open-source GraphRAG Ecosystem Tools empower developers to build more reliable and explainable GenAI applications. By leveraging knowledge graphs, these tools inject rich, structured context into your LLMs to improve response quality and development speed. GraphRAG seamlessly integrates structured and semi-structured data, offering a more comprehensive and reliable foundation for your GenAI endeavors.

Key Highlights:

1. Easy to use - Just load unstructured text to produce a structured graph that surfaces hidden entities and relationships within the data. It works with PDFs, Word documents, YouTube transcripts, Wikipedia pages, and many other kinds of unstructured text

2. Query Your Graph - NeoConverse allows you to interact with your knowledge graph using natural language. It automatically translates your questions into Cypher queries, retrieves relevant information from the Neo4j database, and presents the results in an easily digestible format - either as text or a chart.

3. Build with Your Preferred Tools - Neo4j integrates seamlessly with popular GenAI frameworks like LangChain, LlamaIndex, and Deepset's Haystack, providing flexibility and support for various programming languages and development environments.

Quick Bites

1. Google’s ex-CEO Eric Schmidt just dropped some truth bombs in a recent talk at Stanford. From accusing Google of prioritizing nap time over AI dominance to suggesting students should, ahem, "borrow" content for their next big app, Schmidt's talk was a rollercoaster of hot takes.

   • Google's AI snoozefest? Schmidt claimed Google lost its AI edge because it chose "work-life balance and going home early" over the "work like hell" mentality of hungry startups.

   • Faster but with questionable ethics? He proposed building a TikTok killer by instructing an AI to simply "steal all the users, steal all the music" and then “hire a whole bunch of lawyers to go clean the mess up”  He later clarified he wasn't advocating actual theft... but still!

   • America's power problem? Call Canada (or maybe the Arabs?). Schmidt declared that the US doesn't have the juice to power the future of AI and suggested partnering with Canada for their hydropower. If that fails, maybe the Arabs will foot the bill?

   • AI: Bigger threat than Facebook. Schmidt painted a grim picture of AI's impact, dwarfing even the havoc wreaked by social media and fueling a "trust problem" that could shake the foundations of democracy.

2. The San Francisco City Attorney’s office is suing 16 of the most frequently visited AI-powered “undressing” websites, often used to create nude deepfakes of women and girls without their consent. These websites allow users to upload images of real, fully clothed people, which are then digitally “undressed” with AI tools.

3. OpenAI shut down a covert Iranian influence operation using ChatGPT to spread misinformation on topics including the U.S. presidential election. Though the operation attempted to spread its message across multiple platforms, it gained almost no traction, and OpenAI remains vigilant in preventing further misuse.

Tools of the Trade

1. Traycer: AI Code Review Agent in VS Code that helps you write error-free code. This AI code extension automatically classifies all your code issues into four buckets to give you clear and actionable feedback.

2. Dappier: The world’s first online marketplace for AI content and data rights. Get paid fairly as your licensed content is accessed by AI companies worldwide.

3. Controllable RAG Agent: Advanced RAG solution that uses a deterministic graph to handle complex questions and prevent AI hallucinations. It ensures answers are grounded in the provided data, not pre-trained knowledge.

4. 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

1. Friendly reminder that models trained on 10x more compute than GPT-4 will be released in the next 6 months or so ~
   Matt Shumer
   

2. There has been a rise of 2 AI cults lately
   
   🍓 cult - they go from one rumor to another, constant speculating about mysterious models , imaginary AGI and follow humans pretending to be bots
   
   AI hater cult - they dislike AI and have decided it’s just a huge hype bubble. They claim there is no ROI and gen AI is largely useless and is demoware
   
   Of course, being cults, they both don’t reflect reality ~
   Bindu Reddy

Meme of the Day

That’s all for today! See you tomorrow with more such AI-filled content.

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!

Unwind AI - Twitter | LinkedIn | Instagram | Facebook

Awesome LLM Apps | Sponsor Us

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 (or 20) of your friends!