The CREATELLM framework
Basic LLM operations
Before you can do anything useful in math, of course, you need to understand the basic operations: add, subtract, multiply, divide. The same goes for large language models (LLMs). They have their own set of “moves” - actions like summarizing, expanding, translating, and rephrasing. These are the core functions that let LLMs reshape and remix language.
So, if you want to prompt LLMs well (or build on top of them), we think it’s worth learning their native behaviors. Based on our own experience - as well as some back and forth with ChatGPT - we constructed our own acronym to help distill the basic LLM “operators” into a relatively simple framework - C.R.E.A.T.E.L.L.M.
But first…
5 things to know
📰 Quote: “The newly released findings from Deloitte are based on a March survey of some 600 chief information officers, CTOs, and other tech leaders. In it, 69% of tech leaders said they planned to boost their team's size because of GenAI.” [Business Insider]
▶️ Video: What are evals? | Rachel Woods
🎤 GenAI Prompt: Imagine you took over as the CEO of the company […] What are your goals for growing your company over the next 2-3 years?
🎓 Learning: AI Python for Beginners - DeepLearning.AI. This free course teaches Python programming fundamentals and how to integrate AI tools for data manipulation, analysis, and visualization.
📅 Event: Vibe Coding: Build a Business from Scratch with AI | Girls Who Code | Monday, June 30, 2025 at 7:00 PM ET | Free
C.R.E.A.T.E.L.L.M.
When we first learn math, we’re taught the basic operations: addition, subtraction, multiplication, and division. These operators (and others) are foundational: not just tools for calculation, but a way of understanding how numbers relate, transform, and combine.
Similarly, LLMs operate within their own functional space. They don’t manipulate numbers (at least not primarily); they manipulate language, ideas, and structure. To effectively use, analyze, or even design with LLMs, it helps to know their own core set of operations…their adding and subtracting equivalents.
These operations define what LLMs can do with language - and helps us use these systems more wisely and ask better questions.
The core functional “operations” of LLMs (like math’s +, −, ×, ÷) can be captured by our 9-part framework:
C - Compose
Fusion, synthesis, interpolation
LLMs can blend ideas, complete thoughts, and stitch together fragments into coherent wholes.
R - Recall
Retrieval, hallucination
Retrieve learned patterns or facts; fabricate new ones that “feel” plausible, even if imaginary.
E - Expand
Elaboration, extension, amplification
Given a seed idea, they can grow it - building out scenes, arguments, or explanations.
A - Analyze
Extraction, classification, comparison
Break down input, find structure, make comparisons, and identify what’s important.
T - Transform
Translation, paraphrasing, style shift
LLMs can reshape content across languages, voices, or tones - same meaning, new form. Code creation is a form of translation from common speech (e.g. English) to a programming language.
E - Explain and Execute
Self-reflection, instruction following, justification
LLMs can explain their steps, follow directions - including as AI agents, and reflect on what they just said.
L - Lessen
Summarization, abstraction, compression
Distill long inputs into their core principles or patterns.
L - Logic / Reasoning
Deduction, planning, critical evaluation
They attempt structured reasoning - though often imperfectly - when prompted correctly.
M - Multimodal Mapping
Image ↔ text, audio ↔ text, cross-modal linking
They connect language to other media forms, enabling rich input/output mappings.
We hope CREATELLM helps you better understand what LLMs can do including how they shrink, expand, remix, and reason through language.
Adventure on.