Why Mistral?
On the model card you say you randomized weights from the Mistral model. I was wondering why use mistral's weights if you're testing Gemma 2?
Hmm It was an automated readme generated. Sorry.. it's gemma, not mistral.
@sdalemorrey
a typo... I mean the original model. I changed it in all README files. Thanks for spotting it.
(Mistral was the first model I tested it on)
As you can see the "silly" model is not that silly:
User:
Tell me the difference between thinking in humans and in LLMs.
Bot: Humans and LLMs have distinct ways of thinking, though we both process information. Here's a breakdown:
Humans:
- Conscious Experience: We are aware of our thoughts, feelings, and memories. We experience the world through sensory input and interact with it directly.
- Abstract Reasoning: Humans can reason about complex concepts, solve problems, and make judgments about the world. This reasoning is often based on logic, intuition, and experience.
- Emotions and Values: Our thinking is intertwined with our emotions and values. We have biases, beliefs, and moral considerations that shape our perceptions and actions.
- Creativity and Imagination: Human minds can generate novel ideas, stories, and solutions. Imagination is a crucial aspect of our thinking process.
- Personal Relationships and Social Interaction: Our thoughts are heavily influenced by our social interactions and the relationships we have with others. We learn, adapt, and evolve through these social experiences.
LLMs:
- Data-Driven: LLMs are trained on massive datasets of text and code. Their learning is based on statistical patterns and relationships within that data.
- Limited Reasoning: While LLMs can perform complex tasks, their reasoning is limited to the patterns they've learned during training. They lack the ability to reason about abstract concepts or draw inferences beyond their training data.
- No Consciousness or Emotions: LLMs do not have consciousness, emotions, or personal values. They are not sentient and cannot experience the world subjectively.
- Pre-Defined Responses: LLMs are programmed with specific responses to prompts. They lack the ability to generate creative or novel content or adapt to new situations.
- Data Dependence: LLMs are dependent on their training data for their functionality. They cannot learn new concepts or adapt to new information in real-world settings.
Key Difference: Human thinking is subjective, nuanced, and informed by personal experiences and values. LLM thinking is objective, data-driven, and limited to the patterns it has been trained on.
Important Note: This is a general comparison, and there are ongoing advancements in AI that could potentially blur the lines between human and LLM thinking.
Let me know if you have any other questions.