Week 8 Flashcards
Describe the waves of technological revolution
Roughly every 14 years a technological revolution hits. We are admit a new wave that will change how we work, how we live, and how, hopefully as humanity, we thrive
- 1994
- 2008
- 2022
Currently in the wave of artificial intelligence
Artificial Intelligence
Artificial Intelligence (AI) - computer software that can mimic or improve upon functions that would otherwise require human intelligence
Think of AI as a simulation of human intelligence processes by machines/computer systems that exceed what a human can do alone
Artificial Intelligence can be found in: pattern recognition, medical diagnosis, computer vision, search recognition, self-driving automobiles, natural language processing
What is the ‘fuel’ of Artificial Intelligence
Compute power is the fuel of AI
- AI works more efficiently on specialized chips, designed for rapid computations needed
- Recall: Moore’s Law - compute power doubles approx. every 2 years
- There is a new generation of hardware chips that are tailored to AI use cases (GPU vs CPU)
What are the major types of Artificial Intelligence
Major types of Artificial Intelligence:
(1) Machine Learning
(2) Deep Learning
Now there is a new type called Generative AI
Type of Artificial Intelligence: (1) Machine Learning
Type of Artificial Intelligence: (1) Machine Learning
AI broadly defined as software with the ability to learn or improve without being explicitly programmed
Ex. Netflix uses ML to analyze viewing habits of customers to make prediction around what viewers might enjoy next via the “watch next” feature
There are two SUB-CATEGORIES of machine learning: (1) supervised learning and (2) unsupervised learning
Type of Artificial Intelligence: (2) Deep Learning
Type of Artificial Intelligence: (2) Deep Learning
Sub-category of machine learning and the deep refers to the layers of interconnections of neural networks to arrive at results to process data and make decisions
Ex. Visa/Mastercard uses deep learning to prevent fraud by detecting anomalies in user transactions and creating/modifying user profiles
Supervised Learning
A sub-category of machine learning –> Supervised learning is algorithms trained by specific examples and classifications
Ex. Gmail/Outlook email spam filters learn to classify emails as “spam” or “not spam” by recognizing patterns and features in emails such as certain phrases or sender profiles, which it then uses to classify new, unseen emails
Unsupervised Learning
A sub-category of machine learning –> unsupervised learning is algorithms that are not fed to a pre-determined result
Ex. Facebook’s “People you may know” feature, which uses ‘clustering’ to identify patterns in user data without being told exactly what to look for (i.e., # of connections with people who attend the same school as you)
Generative AI
Generative AI creates new written, visual, and auditory/video content given prompts or existing data
Describe the new generation of hardware chips used for AI
There is a new generation of hardware chips that are tailored to AI use cases (GPU vs CPU)
GPU - graphic processing unit
CPU - central processing unit
Both are essential components in a computer system, however AI tends to favour the parallel processing power of the GPU
- CPU = generalists
- GPU = specialists, designed for specific tasks like handling large blocks of data simultaneously which makes them more efficient for intense computations in AI
- CPUs still required, but handle sequential processing tasks
Consider NVIDIA - one of the biggest winners in this space
Name some popular categories of Software used in AI
(1) Neural Networks
(2) Expert Systems
(3) Algorithms
Neural Networks
Popular categories of software used in AI: Neural Networks - statistical computer model inspired by the human brain
- consist of of interconnected layers of neurons/nodes that process information
- neural networks hunt down and expose patterns, building a multi-layer relationships that humans cant detect on their own
- if a set of interrelationships is strong, they are “approved” in the model
- if a better set of relationships is found, old ones tweaked or discarded
What plays a critical role in Neural Networks
Data plays a critical role and neural networks require a massive amount of data to work
What are some use cases of Neural Networks
Major use case: image recognition and natural language processing
Example: tiktok reccomendations algorithm - by feeding large amo9unts of user data, the algorithm learns and tailors content for each individual user
Expert Systems
Popular Categories of Software used in AI: Expert Systems
Expert Systems - AI systems that leverage set of programmed decision rules or example outcomes to perform a task in a way that mimics applied human expertise
-Take the form of “IF THIS, THEN THAT” decision trees or rules executed by analyzing specific cases against outcomes
-Do X, because Y variable is a certain measure
-Ex. make less product because weather <= 40F
Unlike Neural Networks and other modern machine learning techniques, expert systems do not typically require massive amounts of data to set up.
- However, they do require the ability to extract rules or expertise, which means there is time and expense working with subject matter experts to test and iterate to ensure outcomes are hat is expected
AI vs Algorithm
AI vs Algorithm: set of instructions that tell a computer system what to do
-Not everything is AI, and sometimes is simply an algorithm running in the background
AI USES algorithms as part of its processes, but AI also can modify and improve these algorithms based on the data it encounters
Example: an algorithm might be a recipe for baking a cheesecake. The recipe lists the step by step instructions (the algo) and when followed exactly, with all the inputs, you get the delicious cheesecake
AI starts fixing the algorithm (original recipe) because it learns from the data/experience to adapt and improve results which an algorithm cant do on its own
Generative AI
Generative AI - describes AI that can be used to create new content, including text, audio, images, video, code, and even simulations
Generative AI is a sub-category of machine learning
Machine learning = type of AI that allows models to learn from data patterns without human intervention
This is the breakthrough: move from beyond just perceiving “something is something” and classifying it, we are now at the stage of being able to create based on an input (prompt)
Consider OpenAI’s feature - ChatGPT
Plateau of Productivity - Generative AI
Timeline:
- Innovation Trigger
- Peak of Inflated Expectations
-Trough of Disillusionment
- Slope of Enlightment
- Plateau of Productivity
Plateau of productivity = mainstream adoption starts to take off
Large Language Models (LLM)
Large Language Models - used in AI like ChatGPT, LLMs are specifically trained to generate human like text
- unsupervised models that learn from statistical patterns in language
- at the highest level of abstraction, LLMs are a prediction model
Generative Pre-Trained Transformer (GPT)
Generative Pre-Trained Transformer (GPT) is a specific type of LLM that basically predicts the next word in a sentence based on statistical patterns
- GPT’s don’t actually comprehend the meaning of text put in front of them
-The “secret sauce” is the ability to consider context from both past and future inputs simultaneously, which as a result, enables high quality, contextually relevant output
In LLM’s there is a concept called “_____” which represents a word or part of a word
In LLM’s there is a concept called “Tokens” which represents a word or part of a word
Tokens
Tokens - LLMs process text by breaking it down into tokens which can be words or chinks of characters
Hamburger = “ham” “bur” and “ger” whereas pear = “pear”
Total # of tokens processed in a given request depends on the length of input, output, and request parameters
1 token corresponds to 4 characters of text
ChatGPT3 vs ChatGPT4
ChatGPT3 vs ChatGPT4
- the exponential scale of parameters that these models are being trained upon
- GPT4 has advanced reasoning capabilities that combine with more creative
- GPT3 was more prone to logic and other reasoning errors with more complex prompts
ChatGPT and Platforms
ChatGPT and Platforms:
- OpenAI has launched plugins that will enable ChatGPT to interact with external services built by third party developers as well as web browsing via MSFT BING
- Implications: ChatGPT is becoming a platform (platform businesses allow for the development and integration of software products and other complementary goods, creating an ecosystem of value added offerings)
- Platforms create network effects that create value via exchange, staying power, and complementary benefits
