AI Cheatsheet #1
Let’s kick off by exploring different AI use cases across various departments, and spotlight some key tools and providers in the field.
AI Use Cases
All departments
- Communication: Help draft and correct emails, improving communication effectiveness.
- Knowledge: Read, understand, and summarise documentation.
- Learning: Recommend specific courses and resources to learn from. Productivity: Brainstorm, plan your day
Operations and finance
- Security: Breach risk prediction, respond to incidents
- Finance: Process invoices, detect fraud, assess credit
- Customer support: Answer customer questions
- Admin: Automate E-mail generation, track time
- Legal: Summarise documents, create contracts
- Recruitment (Sourcing): Match CVs, manage onboarding
Business and product
- Product: Generate product ideas, automate note taking
- Strategic Planning: Analyse scenarios, get competitive insights
- Sales: Access predictive sales analytics, manage lead scoring
- Marketing: Generate content, segment target customers
Software engineering
Development
- Suggest code
- Automate code reviews
- Analyse requirements
- Optimise system design
- Predict bugs and security breaches
QA
- Automatically generate test cases
- Prepare visual testing
- Prepare UX Testing
- Identify bottlenecks
- Classify defects
DevOps
- Measure cost-saving
- Recommend resource configurations
- Manage resources
- Set up monitoring & alerts
- Automate workflows
Snapshot of the AI universe
AI experts are innovating in three key areas to improve their quality and efficiency, while trying to evaluate the impact of these changes. Our goal is to help you to understand the most important techniques and keywords in each area.
Context
Prompt with relevant info
Prompt input = Perception
Reasoning
Model and configuration
Trained knowledge + prompt memory = Decision making
Conclusion
Action (give answers, use tools)
Non-deterministic = Same input, different conclusion
Context
It is crucial to give clear context.
Reasoning
Model outputs are affected by the reasoning. You can create and modify this steps through techniques listed below:
Model merging
Mixture of experts
Fine-tuning
Specific training (PEFT, QLora, Lora)
Quantisation
Reduce Size (GPTQ, NF4, GGML)
Configuration
Temperature, number of tokens, Top-K/N, frequency, presence
Conclusion
AI Agents are LLMs interacting with physical or virtual environments and taking complex actions based on conclusions (e.g. send email, message on social media, book a flight). In order to complete their task, they can do several iterations on Context + Reasoning = Action, adding new context in each iteration.
Hallucinations
Non-deterministic seed & fingerprint
Evaluate results
Self-diagnose & self-correct
List of useful links and papers:
Discover some of the resources that feed our research as well as useful links to tools and papers that are fueling advancements in the AI world.
Links:
- AI Community: Huggingface
- GPT Stores: GPTStore
- Prompt Engineering: Strategies and tactics
- AI Toolkit: Langchain | LLama index
Paper:
Impact of AI in the world – 1697521384274
The most positive economic implication of AI disruption will likely be accelerating labor productivity after many years of near stagnation.
AI Cheatsheet #1
If you have questions or require additional context on any of the information shared, please feel free to let us know by using the contact form or sending an email to sayhello@zartis.com.