You look at a model. You see its name. You see its architecture. You see its training data. You wonder: "Where did it come from?" It has a lineage. It has a family tree. It is descended from earlier models. It is a mutation. It is a hybrid. It is a new species. This is the genome of AI. A family tree of models.
The tree is complex. It is branching. It is evolving.
The Roots
The roots are the foundational models.
The Concept:
The first models are the roots.
They are the ancestors.
They are the origin.
The Examples:
GPT-1, GPT-2, GPT-3.
BERT, RoBERTa.
T5, BART.
A Contrarian Take: The Roots Are Not the Beginning. They Are the Foundation.
The roots are not the beginning. They are the foundation. They build on earlier work.
The tree is deeper than we think.
The Branches
The branches are the descendants.
The Concept:
The branches are the descendants.
They are the offshoots.
They are the mutations.
The Examples:
GPT-4, GPT-4o, o1.
Llama, Mistral, Qwen.
Claude, Gemini.
A Contrarian Take: The Branches Are Not Separate. They Are Interconnected.
The branches are not separate. They are interconnected. They borrow from each other.
The tree is a network.
The Mutations
The mutations are the fine-tuned models.
The Concept:
The mutations are the fine-tuned models.
They are the specialized offshoots.
They are the adaptations.
The Examples:
Llama 3 (base), Llama 3 (coding), Llama 3 (medical).
Mistral (base), Mistral (instruct), Mistral (quantized).
GPT-4 (base), GPT-4 (Turbo), GPT-4 (mini).
A Contrarian Take: The Mutations Are Not Separate. They Are Variants.
The mutations are not separate. They are variants. They share the same base.
They are the same species.
The Hybrids
The hybrids are the crossbreeds.
The Concept:
The hybrids combine different lineages.
They are the crossbreeds.
They are the innovations.
The Examples:
Mamba + Transformer.
MoE + SSM.
Open-source + Proprietary.
A Contrarian Take: The Hybrids Are Not New. They Are Recombinant.
The hybrids are not new. They are recombinant. They combine existing elements.
They are a new combination.
The Phylogeny
The phylogeny is the family tree.
The Concept:
The phylogeny is the family tree.
It maps the relationships.
It traces the lineage.
The Components:
Nodes: The models.
Edges: The relationships.
Clades: The groups.
A Contrarian Take: The Phylogeny Is Not Fixed. It Is Dynamic.
The phylogeny is not fixed. It is dynamic. It is evolving.
The tree is growing.
What This Means for You
You are a user of AI. You need to understand the lineage.
- Be Aware:
Be aware of the lineage.
Be aware of the relationships.
- Be Strategic:
Choose the right model for your needs.
Understand its lineage.
- Be Curious:
Explore the family tree.
Understand the evolution.
The Last Genome
The last genome is not a code. It is a history.
You ask: "Where did this model come from?"
The AI says: "It depends."
You realize: The model is not an individual. It is a lineage.
If you could trace the family tree of AI, what would you find? And why?
Top comments (0)