Prompting: It is a call to action, to the large language model.
- LLMS are predictions Engine, its an generic completion which gives statistically best response
We have to Learn to hack probability to obtain best predictive results
1. Personas: give AI some personality (narrow down focus)
2. Context : necessary tool which helps to understand GENAI tools what we need from it
More context = less hallucinations
LLMS are frozen in time : so give it proper context
Give your ai permission to fail (you will explicitly say if its not in the context, you cant find answer, say i dont know) - good fix for hallucination
3. Format: Output requirements (Format, Length, Tone)
We can teach llm to follow a pattern using something called few shot examples or few shot prompting
4. Advance techniques:
A. Chain of thoughts: (showing your work) , step by step answers
Increases Accuracy and Trust
Extended thinking features in all llm
This is when the AI breaks a problem into small steps and reasons through each part. By showing the thinking process, the model becomes more accurate and reduces logical mistakes. It also helps users understand how the answer was formed, which builds trust. Most modern large language models use this internally to handle complex questions, calculations and decisions.
B. Trees of thoughts: Explores (branches) ie multiple paths, which enables ai to do self corrections (generates a diversity of options)
Instead of following one line of reasoning, the AI explores several possible paths at the same time. Each path represents a different way to solve the problem. The model can compare these paths, pick the best one or combine ideas from multiple branches. This gives the AI a chance to correct itself and avoid getting stuck on a wrong assumption.
Two common strategies inside this approach are:
• Playoff method: Competing reasoning paths are evaluated against each other to choose the strongest one.
• Adversarial validation: One path challenges another to expose flaws, which improves the final answer.
This technique leads to more creative, reliable and well-tested solutions.