More Ai Flashcards
(19 cards)
What is Generative AI?
AI systems that can create new content (text, images, code, etc.) by learning patterns from existing data. Unlike traditional AI that classifies or predicts, generative AI creates new outputs.
Use Cases: Content creation (marketing copy, product descriptions), Code generation (GitHub Copilot), Image creation (DALL-E, Midjourney), Text-to-speech (synthetic voices). Business Impact: Can reduce content creation time by 40-70% while maintaining quality.
What is an LLM?
A type of AI model trained on vast amounts of text data to understand and generate human-like text.
Key Players: GPT-4 (OpenAI), Claude (Anthropic), PaLM (Google), Llama 2 (Meta). Technical Detail: Modern LLMs can range from 7B to over 1T parameters.
What is a Transformer in AI?
A revolutionary neural network architecture from 2017 that enables parallel processing of text.
Technical Components: Self-attention mechanism, Positional encoding, Multi-head attention, Feed-forward networks. Why It Matters: Enabled 10x improvement in training efficiency vs. previous approaches.
What are the main types of AI model training?
Three key approaches: Pre-training, Fine-tuning, RLHF (Reinforcement Learning from Human Feedback).
Cost Implications: Pre-training can cost $1M-$10M, fine-tuning starts at $10K.
What is Prompt Engineering?
The art and science of crafting effective inputs to get desired outputs from AI models.
Best Practices: Be specific and detailed, Use examples (few-shot learning), Include context and constraints, Structure complex tasks. ROI: Good prompt engineering can reduce token usage by 30-50%.
What is a Token in LLMs?
The basic unit of text that LLMs process.
Technical Details: Average word = 1.3 tokens, Pricing typically per 1K tokens, Common limits: GPT-4: 8K-32K tokens, Claude: Up to 100K tokens, Llama 2: 4K tokens. Business Impact: Directly affects operating costs.
What is the Context Window?
The maximum amount of text an AI model can consider at once.
Competitive Landscape: Claude: 100K tokens, GPT-4: 32K tokens, Llama 2: 4K tokens. Use Case Impact: Larger windows enable Document analysis, Code review, Complex reasoning tasks.
What are Embeddings?
Numerical representations of text that capture meaning, enabling semantic search and comparison.
Applications: Semantic search, Document clustering, Recommendation systems. Technical Detail: Usually 768-1536 dimensional vectors.
What is RAG?
Technique combining LLMs with custom knowledge bases.
Implementation Methods: Vector databases (Pinecone, Weaviate), Document chunking, Semantic search. Benefits: Reduced hallucination, Custom knowledge integration, Lower training costs.
What are Vector Databases?
Specialized databases for storing and querying embeddings.
Key Players: Pinecone, Weaviate, Milvus. Use Cases: Knowledge management, search, recommendation systems.
What are the main deployment options?
Three primary approaches: API Services, Cloud Deployment, On-premises.
Cost Considerations: API: Pay-per-use, lowest upfront cost; Cloud: More control, medium cost; On-prem: Highest control, highest cost.
What are the key data protection measures?
Multiple layers of protection: Data encryption, Access controls, Audit logging, Data residency options, PII detection and redaction.
Compliance: GDPR, HIPAA, SOC 2.
What safety measures prevent harmful outputs?
Multiple safeguards: Content filtering, Toxicity detection, Bias mitigation, Output validation.
Implementation: Both model-level and application-level controls.
How to compare AI models?
Key metrics: Capability benchmarks, Cost per token, Context window size, Specialization options, Deployment flexibility.
Example: Claude vs. GPT-4: Claude: Longer context, strong reasoning; GPT-4: Higher capability, more integrations.
What components make up GenAI TCO?
Full cost breakdown: Direct Costs, Indirect Costs.
Direct Costs: API/compute costs, Storage costs, Integration development; Indirect Costs: Prompt engineering, Monitoring and optimization, Training and support. ROI Metrics: Cost per task, time saved, error reduction.
What are top enterprise GenAI applications?
Key applications by department: Sales, Support, HR, Legal.
Success Metrics: Time saved, accuracy rates, cost reduction.
How does GenAI adapt to specific industries?
Industry-specific considerations: Healthcare, Financial, Legal, Manufacturing.
Implementation: Usually combines base models with industry-specific RAG.
What’s next in GenAI?
Key developments: Multimodal models, Improved reasoning capabilities, Lower computational requirements, Better factual accuracy.
Timeline: 12-18 month innovation cycles.
How is the GenAI market evolving?
Key trends: Consolidation of providers, Specialized vertical solutions, Open-source advancement, Regulatory framework development.
Market Size: Expected to reach $100B+ by 2025.
