Week 9

Question 1

Autopilot

Answer 1

Autopilot - allowing AI to make all the decisions

Question 2

Co-pilot

Answer 2

Co-pilot: involves AI ASSISTING humans in making informed decisions

Example: Tesla autopilot has the vehicle making all of the decisions without any human intervention, while co-pilot involves the driver being in control of the vehicle, while Tesla’s software features assist in various ways

AI is great at assisting and augmenting human intelligence, but it may never be able to fully replace the human factor

Question 3

Co-pilot era

Answer 3

Many firms are shifting from the autopilot era to co-pilot

The idea is that AI is sitting alongside the user, augmenting and helping to achieve a better outcome, with more efficiency, than without AI

A co-pilot is more than an assistant, its a partner; a partner works alongside offering insights, support, and guidance, complementing and enhancing vs replacement

Question 4

Why treat an LLM as a Co-Pilot

Answer 4

Why treat an LLM as a Co-Pilot:
while the advances in LLMs are impressive we still need to consider that:
- LLMs are imperfect and make mistakes
-LLMs cannot be left unattended
-LLMs are not humans
-LLMs are tools, and like any tool require proper use to get the most of out of them
-LLM outputs should be considered as the starting point, not the end point

Question 5

Why not trust the LLM

Answer 5

• LLMs make mistakes - these errors happen bevause the LLM is reaching beyond its training data or misinterpret a prompt
  recall: ChatGPT relies on statistical probabilities
  example: mary had a little ____ or An apple a day ____

Consider Hallucination and Confabulation

Question 6

Hallucination

Answer 6

Hallucination - perceptions of something that is not actually present in the environment
- Many researchers are rejecting this term for LLM issues because they feel it mispresents that is happening, plus suggests AI has human features

Question 7

Confabulation

Answer 7

Confabulation - when someone’s memory has a gap and the brain convincingly fills in the rest without intending to deceive others

Question 8

Prompt Engineering

Answer 8

Prompt Engineering - optimizing textual input to effectively communicate with large language models

This is powerful: being able to instruct an AI system in nature language democratizes access to these technologies
- However, natural language, particularly English is notoriously imprecise

Becoming a prompt engineer is equal to learning to write clearly and concisely

Question 9

Anatomy of a Good Prompt

Answer 9

Anatomy of a Good Prompt:
- a specific objective (with input)
- a specific format for the output
- a specific list of things to avoid

Question 10

Types of Prompts

Answer 10

Types of Prompts:
- Strategy advice
- Build a referral system
- Improve your writing with feedback
- Enhancing your problem solving skills