LLM
Published
Also known as: Large Language Model, language model, foundation model
Large Language Model — an AI model trained on vast text that can understand and generate human language. Examples: GPT-5, Claude, Gemini.
Source: Common AI/ML terminology
Primary reference ↗A Large Language Model (LLM) is a neural network trained on large text corpora using a next-token prediction or masked language modeling objective. The resulting models develop general-purpose language understanding and generation capabilities that can be steered toward specialized tasks via prompting or fine-tuning.
Scale and Capability
Modern LLMs are characterized by:
- Parameter count: billions to trillions of learned weights
- Training data: hundreds of billions to trillions of tokens from web, books, code, and scientific literature
- Context windows: 128K–1M+ tokens in current frontier models
- Emergent capabilities: reasoning, tool use, code generation — behaviors not explicitly trained for
LLMs in Biological Research
LLMs contribute to biological research workflows in several ways:
| Use Case | Example |
|---|---|
| Natural language → database query | ”Find liver-specific proteins” → structured HPA filter |
| Synthesis and summarization | Aggregating multi-source evidence into a ranked result |
| Biological reasoning | Interpreting conflicts between tissue expression and protein localization data |
| Tool orchestration | Deciding which database tools to call and in what order |
Key Limitations for Scientific Applications
- Hallucination: LLMs can generate plausible-sounding but incorrect biological claims
- Training data cutoff: Knowledge of recently published data requires RAG or fine-tuning
- No native quantitative reasoning: LLMs are poor at arithmetic; numerical validation should use deterministic code, not model inference
- Inconsistency: The same query can produce different results across runs without careful prompt engineering