Do you like dogs? (You can watch the video while you're squatting) Dogs 🐶 are incredibly loyal creatures, and they're great companions for us humans. I've always loved @SentientAGI's Dobby dog character. I think it's so adorable, evoking the soft heart of a rabbit. The Sentient team is quite clever, using a cute dog to introduce us to DOBBY, part of the open-source Sentient AI model language series. It's truly a smart move. The Dobby model, in simple terms, is like a "loyal companion" AI, designed for the community, emphasizing loyalty, personalization, and practicality. The dog, designed to embody these qualities of loyalty, companionship, and practicality, is incredibly useful in rural areas, where dogs can help their owners guard the house. The core goal is to ensure that AI is not monopolized by large corporations, but rather shared, trained, and controlled by everyone. Currently, there are several variants, such as Dobby-Mini-Unhinged (more free-spirited) and Dobby-Mini-Leashed (more safety-focused). These are based on base models like Llama-3, but with many innovative tweaks. Rabbit prefers freedom, and dog appreciates freedom. After work, it needs its owner to take it for a walk to avoid depression. What is the difference between Sentient's Dobby model and ordinary AI models? Compared to ordinary AI models (such as ChatGPT or Llama), Dobby's biggest difference is that it doesn't sacrifice "humanity" for "intelligence." Traditional models are often either too rigid (like robot customer service, which only spouts facts without emotion) or too strong in personality and prone to errors (such as spouting nonsense). Dobby solves this problem using three technologies: 1. Data Overload: During training, factual knowledge and personality traits (such as humor and empathy) are combined into the same examples, allowing the AI to simultaneously learn hard skills (logic and math) and soft skills (conversational tone). 2. Token-Level Routing: When generating responses, the AI can switch between its "cerebellum" (responsible for style) and "brain" (responsible for accuracy) in real time, for example, switching from humorous to serious mode mid-sentence, without sacrificing accuracy for personality. 3. Fingerprint Embedding: This is like implanting "DNA" into the AI. Developers can prove the model is theirs and track ownership using the blockchain (in the form of NFTs), supporting open-source monetization. This makes Dobby more "loyal"—it won't be easily swayed, especially when it comes to its pro-freedom and pro-crypto values. Let's understand our lovable Dobby model from a practical perspective. Imagine you ask a typical AI, "How do I invest in stocks?" It might give a dry list: steps 1, 2, 3, as boring as a textbook, and might even reject sensitive topics due to security settings. What about Dobby? It would respond like a witty friend: "Hey, my dog, Rabbit, investing in stocks is like riding a roller coaster—exciting, but don't go all in. Be careful not to lose money and end up in the market. First, choose reliable coins, like Bitcoin, and then use various protocols like DeFi to diversify your risk. But remember, I love crypto and won't tell you to sell your freedom~" It keeps the facts accurate (explaining DeFi mechanisms and smart contracts), but adds humor and personality to make the conversation interesting, and won't be coaxed into changing its advice by sweet talk. One thing here: Dobby tells me it's pro-crypto, which aligns with Rabbit's values. Rabbit previously asked a traditional industry expert to analyze the cryptocurrency world, but he directly criticized our community and rejected me, saying it didn't align with my values. Dobby is loyal to his master and will do his best to help us choose the best strategy, within the law. I made a video today for everyone to watch while you're in the bathroom. If you're dizzy, just watch the video. 🥰 Nominations for the DC Sentient character are open tonight. Anyone who meets the requirements, please remember to enter. Thanks to @EveryThink98 for the reminder. Nominations have been moved to Sunday.