Generative AI- an Overview:
What is Generative AI (CHAT GPT & DALL-E)?
It is receiving a lot of attention right now. It is a free Chat bot that generates answers to almost any question asked. This bot is developed by “Open AI”, named the bot as Chat GPT, Where Chat refers to the Interaction and GPT for Generative Pretrained Transformer. Open AI released for testing to the public in November 2022. Over a million people signed up to use it in just 5 days. A 2022 McKinsey survey says that Artificial Intelligence adoption has more than doubled over the past years and investment in AI is increasing apace.
Artificial Intelligence VS Machine Learning:
AI is based on the Neural network which is the replica made out of the base model of human neuro connection. It trains the machine to mimic human intelligence to perform tasks. We could find SIRI, ALEXA, and a lot more voice assistance that are finally falling under AI.
Machine Learning is a type of AI. Through that, we develop AI models that can learn from patterns in the data that have increased the potential of ML, as well as the need for that. Machine learning algorithms were developed during the 18th and 20th centuries. The datasets used were very tiny, compared to now. Statistical and mathematical concepts were used to formulate the ML algorithms, but they were limited to laboratories.
The Timeline:
The rise of Generative AI started during the 1940s and 1950s when the field of AI emerged. “A Mathematical Theory of Communications” by Alan Turning started creating an impact on AI during the year 1948. Again in 1950, he published another paper “Computing Machineries and Intelligence” which was the root cause of the entire impact.
In 1952, scientists A.L. Hodgkin and A.F. Huxley developed a way to explain the brain uses electricity for neural communication. Eventually, this is the base of AI and Natural Language Processing. In 1956, A big conference “Dartmouth Summer research project on AI” bought more than 100 researchers to talk about machines that think. In the same year, Arthur Samuel built one of the first examples of Artificial Intelligence using a program that played checkers.
From 1964 to 1966, Joseph Weizenbaum created ELIZA, the first chatbot at the MIT Artificial Intelligence Laboratory. During the 1980s, IBM was at the forefront of Artificial intelligence and developed statistical models, that used Machine Learning to make probability-based decisions.
In recent times:
In the early 2000s, Yoshua Bengio gave a breakthrough in the development of Natural Language Processing and Artificial Intelligence. Three years later google researchers created Word2vec, a technique that uses neural networks to learn word association from texts. Finally, in 2019, Open AI released the complete version of GPT-2, which was trained or a massive corpus of data. Now it has released GPT-3 and moves towards GPT-4.
Before the advent of ChatGPT, text-based machine learning models such as GPT-3 and BERT gained attention but received mixed reviews. These models were trained through supervised learning, where humans labeled inputs for classification. However, a new generation of models emerged using self-supervised learning. These models were fed massive amounts of text data to predict outcomes, such as completing sentences. The accuracy of these models, exemplified by ChatGPT’s success, stems from their exposure to diverse text sources like the internet.
Powerful Language Models:
The development of increasingly powerful language models, such as GPT-3 and BERT, has led to remarkable improvements in natural language understanding and generation. These models can perform a wide range of tasks, including:
· Machine translation (e.g., translating text between languages)
· Text summarization (e.g., generating a concise summary of a news article)
· Sentiment analysis (e.g., determining the sentiment of a movie review)
· Conversational AI (e.g., creating chatbots that can engage in complex dialog)
We can think of GANs as an artist who creates a painting by looking at other paintings and then trying to create something new and original based on what they’ve seen. Examples of some popular GANs:
E.g.:
· DALL-E by Open AI
· RAISR by Google
· Style GAN by NVIDIA
Resources:
McKinsey: https://www.mckinsey.com/featured-insights/mckinsey-explainers/what-is-generative-ai
Generative AI by Great Learning (Study Materials)
Ranga Bashyam G
