Pharma AI glossary
Large language model (LLM)
A deep-learning model trained on vast text corpora to predict likely next tokens, from which capabilities in drafting, summarization, reasoning-like behavior, and code emerge.
Definition
A deep-learning model trained on vast text corpora to predict likely next tokens, from which capabilities in drafting, summarization, reasoning-like behavior, and code emerge. Strengths: fluency, breadth, speed. Structural weaknesses: hallucination, non-determinism under default sampling, arithmetic unreliability, knowledge cutoffs. Regulatory status in pharma: excluded from critical GMP applications (draft Annex 22); usable with human verification for documentation work; FDA imposes no categorical exclusion but the credibility burden for a high-influence LLM use would be formidable.
What LLMs actually do
LLMs predict the next token from patterns in training data. They excel at language, summarization, and drafting, not at guaranteed factual recall. In pharma they usually sit inside an approved agent or application with retrieval, access control, and logging, not as a standalone browser tab on the manufacturing floor.
Regulated use requires boundaries
Draft EU GMP Annex 22 excludes generative AI/LLMs from critical GMP applications as drafted; FDA accepts performance-based credibility if context of use matches risk. Whether the use case is deviation support, batch record summarization, or SOP search, the question is whether the deployed harness has governance (approved use case, data boundaries), workflow (human review before official commit), and auditability (attributed prompts, sources, and actions).
Frequently asked questions
Can AI support deviation investigations?
It can assist drafting and research when wrapped in an approved agent harness. Governance defines the use case and data boundaries; workflow requires qualified review before anything enters QMS; auditability logs retrieval, tool calls, model version, and human sign-off. The model is one component; inspectors care about the controlled system around it.
Where can LLMs assist other GMP workflows?
Typical governed uses include SOP and batch record search, CAPA and change-control drafting for review, training and onboarding summaries, and submission-support writing in non-critical paths. ML-driven manufacturing tools (vision, sensors) usually sit outside the LLM stack but still need literacy on probabilistic outputs and monitoring.
Why do LLMs hallucinate?
They optimize for plausible language, not verified truth. Retrieval grounding, citations, and human review are engineering and process controls, not optional polish.
Related terms
Benchmark your AI literacy
Free 20-question assessment, eight modules, and instant scoring, built for pharma quality and manufacturing teams.
Open the Literacy CenterBrowse all terms
See the full 71-term glossary index or search interactively in the Literacy Center.
Back to glossary hub →Educational content only, not legal or regulatory advice. Regulatory guidance cited here includes drafts (FDA AI credibility guidance; EU GMP Annex 22) as of August 2026; verify against final texts before relying on them in submissions. Company-reported figures (Merck CSR timings, Sanofi results) are labeled where used. MIT's ~95% pilot figure carries its own caveat (~150 interviews, contested definitions, not peer-reviewed). Re-check sources on module finalization.