Introducing BlindChat: The AI Chatbot That Hides ALL Your Secrets!

So there's a new open source project designed to build a fully interactive and private conversational AI. It's named Blind Chat and I find it truly impressive. In this video, I will discuss what Blind Chat is, its functioning, its significance for privacy, and ways you can incorporate it into your applications. So let's get started. Okay. Blind Chat is a privacy first in browser AI chat system that lets you have natural conversations with large language models without sending any of your data to the cloud. Unlike other chat systems where what you say is sent to the company's servers, Blind Chat keeps everything on your device. For example, when you talk to Google Assistant, Siri or Alexa or even ChatGPT, they send your conversations to their servers and you don't really know what happens to that information. You're basically giving away your personal information, your preferences, your opinions, your secrets, and anything else you might say to the AI provider. Blind Chat is different. It's it doesn't send your chat data anywhere. Everything you type or say stays on your device. Nobody else, including Blind Chat's creators, can see or get your data, which gives you full privacy. And you can still chat smoothly with the AI on Blind Chat, ask questions, and discuss various topics. Now, Blind Chat uses two main techniques to ensure privacy, Local inference and secure enclaves. Local inference means that the model runs on Your device using JavaScript and web assembly technologies. These are standard web technologies that allow you to run complex computations in your browser without installing any software or plugins. Blind Chat uses these technologies to download a pre trained LLM model from a public repository and run it on your device using your browser's resources. This way you don't need to send any data to the server to get a response from the model. If you want to use a big model like GPT3 or Flan T5, but it's too large for your device, secure enclaves can help. These are special hardware features that allow you to run code in a protected environment that is isolated from the rest of the system. This ensures that even if someone accesses the server, they can't interfere with the program or data in the enclave. Blind Chat uses these enclaves to safely operate big models on trusted servers. Your data is changed to code before leaving your device and is only turned back to its original form inside the enclave. The model operates within the enclave and sends you a coded answer that only your device can read. So you can use any model without fearing data exposure or threats. The blindchat website has a user friendly interface letting you type and communicate with the model. There's an option to process your input either directly on your device or on a remote system. If you use the on device option, the model processes your messages without sending them to a remote server. Installation Ensuring data privacy. You can utilize it for multiple purposes, whether for learning, entertainment, research or professional tasks, all while ensuring your data remains confidential and secure. It is actually great for both regular users and developers. Regular users who want to talk with conversational AI but also value their privacy will like Blind Chat. It lets them pick how they want to chat. They can choose local inference, which is super private but but might not be as powerful. Or they can use remote Enclave mode, which uses bigger AI models but means they need to trust the server they're using. Either way, they get more privacy than with other chat AIs. Developers who want to make their own chat AIs using LLMs will also find Blind Chat useful. It lets them use any LLM model like ones from TF Hub or their own creations. Plus, Blind Chat has an easy API. This means developers can easily link their software to Blind Chat, send and get data from the LLMs, and adjust how the models work. How does Blind Chat work under the hood? Well, it has a unique infrastructure that consists of three main components, the ui, the private LLM, and the storage. So when you open Blind Chat on your browser, what you see is called the ui. It shows you where to type, read messages, and lets you choose how you want the chat to work. This UI also keeps your data safe by changing it to a secret code before sending or getting it from the private LLM. Now, the heart of Blind Chat is the Private LLM that runs either on your device or on a trusted server inside a secure enclave. It receives encrypted data from the ui, processes it according to the task instructions, and sends back encrypted results to the ui. The private LLM can be any model that supports text to text format such as T5, Flan, T5, GPT3, etc. Lastly, there's storage, which is the component that stores the LLM models and their metadata. The storage can be either a public repository or a private server, depending on the source and ownership of the models. It provides access to the models for downloading or running them on demand. Now, Blind Chat has some exciting features on the horizon that will make it even more powerful and useful. One new thing coming is connectors for pulling data from various sources such as databases, APIs, files, etc. This will allow you to use external data in your conversations with the LLM models. They're also teaming up with Llama Index ts, a tool for LLM apps. It helps you bring in, organize and use your personal or specific data when chatting with LLM models. Plus, they're launching models that can handle multimodal inputs and outputs such as images, audio, video, etc. This will allow you to have richer and more diverse conversations with the LLM models. Now, how does Blind Chat stand against other chat AI programs? There are three main on device prediction, regular AI APIs and zero trust AI APIs on device prediction means the AI works entirely on your device, no need for the Internet or a server. Blind Chat does this in its local inference mode. This way you get a lot of privacy and it works offline. But it might not be as powerful or fast because it depends on your device's power. Also, you need to download the model first, which can take up space and time. Regular AI APIs run on a distant server managed by big companies like Google or Amazon. This is how most chat AI programs work. With this, you can use strong AI models that can do many things, but you send your data to these servers which might not be private or safe. Plus, you need a good Internet connection and hope the server works well or else there might be delays. 0Trust AI APIs work on a secure distance server using technologies like Intel SGX and or AMD cv. Blind Chat does this in its remote enclave mode. This way you can use strong AI models and still keep your data private and safe. However, you still need the Internet and to trust the server provider. Also, encrypting and decrypting data might make things a bit more complex. So as you can see, Blind Chat offers a unique and innovative solution for conversational AI that balances privacy and performance. You can pick how it works based on what you want and use any LLM model you like. It's a big step forward for conversational AI. Where is it going next? It's still being worked on, but it has a lot of room to get better. The developers of Blind Chat have a roadmap that outlines their vision and goals for the project. Some of the things they plan to do are Add more LLM models to their repository. Improve the privacy and security of their services using techniques like homomorphic encryption, differential Privacy, Federated Learning, etc. Add more features to their API like voice interaction, understanding feelings and detecting emotions. Create more applications and integrations using Blind Chat such as chatbots, voice assistants, content generators, etc. Blind Chat is an open project and anyone who cares about conversational AI and privacy can help out. If you're curious. Their code and other details are on their GitHub page and the links are in the description of this video. Alright, so I hope you had fun getting the scoop on Blind Chat and how it rolls. I reckon it's a super cool and major project that should totally get more eyes on it. If you're with me on this, go ahead and hit that like button and show it to your friends and hit subscribe and ring that bell so you know when I drop something new. Thank you for watching and I'll catch you in the next.