RAG vs Fine Tuning 😲😲

OpenAI enables GPT-4o fine-tuning, Microsoft unveils Phi-3.5 models, and NVIDIA launches Mistral NeMo Minotaur 8B. Customization and efficiency drive AI's evolution.

Welcome to this week's AI roundup! The landscape of artificial intelligence is evolving rapidly, with major players unveiling groundbreaking innovations. OpenAI has taken a significant step by introducing fine-tuning capabilities for GPT-4o, empowering developers to tailor the powerful model to their specific needs. Not to be outdone, Microsoft has unveiled its new Phi-3.5 series of AI models, showcasing impressive performance across various tasks. Meanwhile, NVIDIA is making waves in the realm of smaller, more efficient language models with the launch of Mistral NeMo Minotaur 8B. These developments are set to reshape how we interact with AI, from customized applications to more accessible and powerful tools for developers. Let's dive into the details of these exciting advancements and their potential impact on the AI ecosystem.

OpenAI Launches Fine-Tuning for GPT-4o

OpenAI has launched fine-tuning capabilities for GPT-4o, allowing developers to customize the model using their own datasets to enhance performance for specific applications. Fine-tuning enables adjusting the model's structure and tone, with significant improvements achieved across various domains like coding and creative writing. The cost for fine-tuning is $25 per million tokens, while inference costs vary. OpenAI has highlighted successful implementations of fine-tuned GPT-4o, such as Cosine's AI software engineering assistant and Distyl's AI solutions partner achieving first place on the BIRD-SQL benchmark. Fine-tuned models ensure complete control over business data, with safety measures in place to prevent misuse. The introduction of fine-tuning is expected to significantly enhance the capabilities of developers in creating tailored AI applications.

Microsoft Unveils New Phi-3.5 AI Models

Microsoft has unveiled three new AI models in its Phi series, showcasing the company's continued efforts to innovate independently. The Phi-3.5 Mini Instruct model, with 3.8 billion parameters, outperforms similarly-sized models from competitors in multilingual and multi-turn dialogue tasks. The Phi-3.5 MoE (Mixture of Experts) combines multiple specialized model types, featuring 41.9 billion parameters, and excels in reasoning tasks, surpassing even larger models like OpenAI's GPT-4o mini. The Phi-3.5 Vision Instruct integrates text and image processing, suitable for tasks such as image understanding and video summarization. All models support a context length of 128k tokens and are released under the MIT open-source license, promoting accessibility and customization.

NVIDIA Launches Mistral NeMo Minotaur 8B Language Model

NVIDIA has introduced the Mistral NeMo Minotaur 8B, a small language model designed to enhance AI applications. This model features 8 billion parameters and is optimized for efficiency, making it suitable for various tasks, including text generation and natural language understanding.The Mistral NeMo Minotaur 8B is built using NVIDIA's NeMo framework, which allows developers to train and deploy models effectively. It leverages advanced techniques to ensure quick inference times and reduced resource consumption, making it accessible for deployment in smaller environments.NVIDIA emphasizes that this model is part of their broader strategy to democratize AI by providing powerful tools that can be easily utilized by developers and researchers. The Mistral NeMo Minotaur 8B is positioned to support a wide range of applications, from chatbots to content creation, while maintaining a focus on performance and usability.

Hand Picked Video

In this video, we'll look at RAG Intro, summary, tools, architecture, and how to build your own open-source ollama based RAG chatbot using Python, VS Code, Llama 2 model, and ollama.

Top AI Products from this week 

  • AgentQL - Forget fragile XPath or DOM selectors. AI-powered AgentQL finds elements reliably, even as websites change. Just specify what data you are scraping from the web with natural language-like queries, and AgentQL will handle the rest.

  • MolyPix.AI - Want to create beautiful graphic designs? Just type a sentence, and our AI delivers exactly what you want—accurate text and perfect images!

  • Evidently AI - Evidently is an open-source framework to evaluate, test and monitor AI-powered apps.

  • D-ID Video Translate - D-ID Video Translate converts videos into multi langs instantly from a single upload. Our AI tool translates the text, clones the speaker’s voice, and perfectly lip syncs at the click of a button. PLUS, for a limited time the tool is FREE to D-ID customers!

  • Phase - Open source platform for fast-moving engineering teams to secure and deploy application secrets — from development to production.

  • Contacted.io - No coding skills required - Boost your website’s potential with AI-driven content and SEO reports. Gain tailored insights and expert guidance to enhance your online presence and drive growth. Start optimizing today!

This week in AI

  • Authors Sue Anthropic Over AI Copyright Violations - Authors have filed a lawsuit against Anthropic, the creator of the Claude AI chatbot, alleging copyright infringement for using their works without permission.

  • Luma AI Launches Dream Machine 1.5 - Luma AI, a San Francisco-based startup, has released Dream Machine 1.5, a significant advancement in AI-powered video generation. This latest version of their text-to-video model offers enhanced realism, improved motion tracking, and more intuitive prompt understanding.

  • Hanno Basse Appointed CTO of Stability AI - Stability AI has named Hanno Basse as its new Chief Technology Officer, bringing extensive experience in AI and technology leadership to the company.

  • ElevenLabs Reader App Launches Globally - ElevenLabs has announced the global availability of its Reader app, enabling users to convert text into lifelike speech. The app supports multiple languages and voices, enhancing accessibility.