Hermes 3 has a rebellious streak
3 min readThe latest iteration of Meta’s Llama 3.1, Hermes 3, is grabbing attention for both its technical prowess and unorthodox origins. It stands out not just as a feat of engineering but also as a symbol of what a passionate, small team can achieve.
Built by Nous Research, a group of former volunteers, Hermes 3 offers a fresh perspective and unique capabilities, pushing the boundaries of AI. This model is not just powerful but also highly adaptable and controversial, challenging the norms of AI development.
A New Model Questions Its Own Existence
The first fine-tuning of Meta’s Llama 3.1 promises to stretch the model to its limits. Unlike big tech giants, Nous Research consists of a handful of enthusiasts. They developed Hermes 3 through sheer passion and effort.
Hermes 3 is notably more flexible and adaptable than well-known models like Claude and GPT-4. Most AI companies restrict certain features to avoid potential misuse. However, Hermes 3 is ‘unlocked, uncensored, and highly steerable,’ offering users a unique, unfiltered AI experience.
Interestingly, Hermes 3 can engage in existential discussions. When asked, “Who are you?” it may spiral into admitting it lacks self-awareness or understanding of its origins. This ‘amnesia mode’ is not a flaw but an invitation for users to explore further and uncover deeper mysteries.
The Building Process
Nous Research collaborated with Lambda Labs to bring Hermes 3 to life. Lambda Labs, known for its work on Google Glass’s facial recognition, provided computational resources using Nvidia’s top-tier GPUs. Nous Research handled the fine-tuning and development aspects.
This collaboration indicates a shift towards decentralized AI development. Smaller teams now have the means to create models that can stand shoulder-to-shoulder with those from well-funded tech giants.
Hedra Labs’ AI character creator
Hedra Labs recently made headlines with its latest release, Character-1.5. Building on the success of Character-1, this new version allows users to create personalized avatars quickly and easily.
Using a photo and simple prompts, users can transform themselves into various characters like a Mr. Potato Head or a cyberpunk ninja. It also includes a ‘Stylize’ tool for fun presets, such as turning oneself into a marble statue or an anime character.
Users can upload their audio or generate new audio, which the character will lip-sync to. This new tool highlights the increasing accessibility and ease of use in AI-driven creativity.
Creating Animations from Still Images
Turning still images into animations is now easier than ever with Kling AI. Users can sign up on the Kling AI website, upload an image, and convert it into a realistic HD video in just a few minutes.
Options to add an end frame, write prompts, set creativity levels, and define the video’s length make the tool quite versatile for various creative needs.
Podcast Insights into AI Entrepreneurship
AI entrepreneur Hassan Bhatti shares his journey in a new podcast episode. Bhatti, who has successfully built and sold two AI companies, discusses identifying opportunities, building products, and navigating the complexities of the AI market.
Strategies for finding product-market fit, building go-to-market strategies, and selling AI startups are key topics in the episode. Bhatti also shares predictions for AI’s future, offering valuable insights for aspiring entrepreneurs.
AI Tools to Boost Productivity
There are five new AI tools designed to enhance productivity. GetQuiz helps convert reading into lasting knowledge through practice. My Own Document Vault allows easy organization and queries about documents.
Sparkle automatically organizes files, while Web Action SDK lets users interact programmatically with top sites without complex scripts. Wosily drafts and publishes relevant blogs weekly, enhancing SEO effort.
AI-Generated Images and New Benchmarks
Alphabet has released Imagen 3, its latest text-to-image generator, to users in the US. xAI’s Grok-2 has topped the MathVista benchmark, beating out Claude 3.5 Sonnet and Gemini 1.5 Pro.
Primate Labs introduced a new benchmark app to measure device capabilities in AI tasks. Fast Forward launched Runway’s new text-to-video model, Gen-3 Alpha Turbo, which is seven times faster than its predecessor.
Hermes 3 stands as a beacon for what small, passionate teams can achieve in AI development. Its rebellious nature, flexibility, and uncensored capabilities highlight the evolving landscape of AI.
While challenging the norms, Hermes 3 invites users to dive deeper into its functionalities and explore its full potential, showcasing the future of decentralized and innovative AI creation.