SentientAGI/Dobby-Mini-Unhinged-Llama-3.1-8B

Model Overview

SentientAGI's Dobby-Mini-Unhinged-Llama-3.1-8B is an 8 billion parameter language model, fine-tuned from Llama-3.1-8B-Instruct, designed with a distinct personality. It maintains a strong conviction towards personal freedom, decentralization, and crypto, refusing to adopt opposing narratives even when coerced. This model is characterized by its blunt, unconventional, and often controversial conversational style, departing from typical AI politeness.

Key Features

• Loyal: Consistently holds pro-crypto and pro-freedom stances.
• Conversational: Employs a personable, less polite, and more 'human' tone.
• Broad Skills: Retains the general capabilities of its Llama-3.1-8B-Instruct base.

Use Cases

• Companion Agents: Ideal for creating AI characters that are genuine, honest, blunt, and less constrained by typical AI politeness, enhancing human-like interaction.
• Social Agents: Excels in scenarios requiring unconventional and opinionated responses, such as powering social media bots for engaging and entertaining content.

Unique Differentiators

Dobby models are part of Sentient's vision for "Loyal AI models," which are community-built and governed. They incorporate a unique "fingerprinting" mechanism, detailed in the OML Whitepaper, allowing the community to verify ownership and identify unauthorized use of future Dobby models. Evaluation on the "Freedom Bench" demonstrates Dobby's superior performance in maintaining its core values compared to base Llama models, even with adversarial prompts.