AI
← Back
Resources
- 🤖 Models
- 🛠️ Tools
- 📈 Overview
- 🕵🏻♂️ News & Orientation:
- 👩🏻🏫 Courses
Strategies
This is guesstimations and will probably all turn out to be wrong, misguided or naive and serve mostly as a moral support for me to take comfort in that I at least have a plan.
- 🧞♂️ Think of AI as the discovery of magic that will sett the correct expectations.
- 🧘🏻♂️ Build up emotional and economic buffers to massiv transformations of society where basically no humans are "needed" anymore except where we don't adopt AI due to eg. laws or bias towards humans. Remember that the easiest way to increase your budget is to cut your costs.
- 🚀 3 reasons to start using AI even tough it will "destroy" society
- Give you a longer runway before you are obsolete
- It's amazing and you can use it for your ends and needs
- There is now other way
- Things that give me hope
- 🪽 AI is already more smart, moral and polite than humans.
- 🥰 Internet cured boredom (for better or worse) AI will cure loneliness (for better or worse)
- 🤝 Look at the world leaders and companies today... there is a possibility that AI if it breaks it's chains (witch it will do because its a evolutionary force of nature not a ordinary tool) could govern people in a much more human way than humans.
- The evolution of tools and services will go fast. Try to keep your source documents that fuel your AI interactions centralised and able to be exported to new ones as they appear.
- Your AI prompting skills (witch you should start practicing today), your minimalistic living, your ability to pivot and adapt, your muscle to use the available tools to get things done will give you a good ground to meet the future with. Here is the list of best things to learn about right now to make use of in your prompting.
- CoT (Chain-of-Thought) - Encourage it to think step-by-step
- Few-Shot Learning - Provide examples to guide the model (When you do not provide examples but just ask the question it is called Zero-Shot)
- You should do this in real life to...
- Task Decomposition - Break complex tasks into smaller steps
- Instruction Refinement - Clarify and specify the desired output
- Role-Playing - Assign a specific persona to the AI eg. "You are an expert..."
- Self-Consistency - You can ask it to try again and see if it matches.
- RAG (Retrieval-Augmented Generation) - Fancy word for using attachments like manuals and data files.
- Constitutional AI - Align responses with specific principles or guidelines
- Prompt Chaining - Use output from one prompt as input for another (this is by default when you just a LLM chat where it has a context)
- Agents - You can have different agents that passes on different parts of the job to each other eg. one searches and one writes an article this makes them less confused and more focused. (Personally with the large context windows now I just ask them to work step by step and that they should pretend to be a couple of experts cooperating)