Lee Harrold

GenAI Capabilities Reference Sheet

1 Core Text Understanding & Generation

Base LLMs

Description: Foundation models for understanding and generating human language

When to use: General text generation, understanding, and dialogue

Good at:

Bad at:

Text Embeddings

Description: Convert text into numerical vectors capturing semantic meaning

When to use: Semantic search, clustering, similarity detection

Good at:

Bad at:

Chain-of-Thought Prompting

Description: Guide LLMs to show step-by-step reasoning

When to use: Complex problem-solving, debugging, explanation generation

Good at:

Bad at:

Few-Shot Learning

Description: Train models with minimal examples in the prompt

When to use: Quick adaptation to new tasks, formatting consistency

Good at:

Bad at:

Zero-shot Classification

Description: Categorize content without prior examples

When to use: Dynamic categorization, flexible classification needs

Good at:

Bad at:

2 Knowledge & Retrieval

RAG (Retrieval-Augmented Generation)

Description: Enhance LLM responses with external knowledge

When to use: Need accurate, up-to-date, or domain-specific responses

Good at:

Description: Advanced search understanding meaning, not just keywords

When to use: Information retrieval, document search, content discovery

Good at:

Text Summarization

Description: Create concise versions of longer texts

When to use: Document processing, content briefs, information extraction

Good at:

Content Generation with Control

Description: Generate text with specific parameters (temperature, sampling)

When to use: Need specific creativity/randomness levels

Good at:

3 Multimodal Capabilities

Vision Language Models

Description: Process and understand images alongside text

When to use: Image analysis, visual Q&A, content moderation

Good at:

Image Generation

Description: Create, edit, or modify images from text descriptions

When to use: Design ideation, content creation, image editing

Good at:

TTS (Text-to-Speech)

Description: Convert written text into natural-sounding speech

When to use: Accessibility features, audio content, voice assistants

Good at:

STT (Speech-to-Text)

Description: Convert spoken audio into written text

When to use: Transcription, voice commands, accessibility

Good at:

4 Developer Tools & Code

Function Calling

Description: Parse input into structured API calls

When to use: Building AI assistants with external services

Good at:

Structured Output

Description: Generate responses in specific formats

When to use: Data extraction, API integration, format conversion

Good at:

Code Generation/Completion

Description: Generate or suggest code completions

When to use: Development assistance, code conversion, documentation

Good at:

Model Optimization

Description: Optimize models for production deployment

When to use: Production deployment, resource constraints

Good at:

Bad at:

5 Customization & Optimization

Fine-tuning/PEFT

Description: Customize models for specific tasks efficiently

When to use: Domain adaptation, specific task optimization

Good at:

Bad at:

Knowledge Distillation

Description: Create smaller models learning from larger ones

When to use: Model compression, deployment optimization

Good at:

Bad at:

6 Orchestration & Control

LLMs as Agents

Description: Autonomous LLMs for multi-step tasks

When to use: Complex task automation, workflow orchestration

Good at:

Bad at:

Prompt Chaining/Orchestration

Description: Coordinate multiple AI tasks in sequence

When to use: Complex workflows, multi-step processing

Good at:

Bad at:

Ranking Models

Description: Score and order items by relevance/quality

When to use: Search optimization, content curation

Good at:

Bad at:

Self-Reflection/Self-Correction

Description: Models that can evaluate and improve their own outputs

When to use: Quality assurance, iterative improvement

Good at:

Bad at:

LLM-as-Judge

Description: Use LLMs to evaluate outputs/solutions

When to use: Quality assessment, content moderation

Good at:

Bad at: