As we all embark our journey towards using Artificial Intelligent systems, the building block we stumble upon is ‘Generative AI’. For someone new to AI it is important to understand the new paradigm and this blog take us towards some basic knowledge for beginners. The following sections just browses over some Myth ‘n’ Might(noun) of AI helping us quick adaption.

Myth 1.

ChatGPT filters information to provide only relevant. I better ChatGPT than googling…

Search engine v/s Generative AI (chatbot – pre-trained models)

• Search engine looks for data in the internet/indexed data store whereas GenAI generates ‘new’ data based on the training.
• Search engine provides existing data reference/location whereas GenAI data is absolute and will not be hosted or available for access in internet.

Summary:

Generative AI is positioned to provide information based on the request details (and pre-trained information), for a specific purpose. Search engines just browse the internet and provide public hosted accessible links.  Purpose of searching is completely different than creating/generating information.

Might 1.

Large Language Model (LLMs)

Large Language Models(LLMs) are Machine Learning (ML) models that are trained using massive textual data and then fine-tuned for generating new content. In LLMs, Deep learning algorithms are used for understanding natural language.

• LLMs can generate textual content in language the model is trained.
• LLMs can understand any natural language only if the same is used in training.
• LLMs can perform language tasks, such as translating languages, sentiments, conversations, and etc.,
• LLMs can process complex textual data, entity identification and entity relationship building; then generate new textual content that is coherent and grammatically correct.

Example: ChatGPT / GPT3 – (Generative Pre-trained Transformer 3) facts :

• Trained using 175 billion parameters, 3.2 M batch size and 96 attention layers.
• Trained using 45TB of data in different sets.
• Trained for 355 GPU years, considering Tesla V100.

Myth 2

AI First is like Mobile First where users are presented with Gen-AI solutions.

AI First Enterprise v/s Mobile First Solutions

• AI First software engineering can be a process/methodology to follow while implementing solutions whereas Mobile First is experience design approach of providing solution to end users.
• AI First focus on decisions based on data for making sure the users are provided personalised experience whereas Mobile First focus on presenting data to users on mobile devices (mostly all variants) as priority along with other presentation options like web.

Summary:

Generative AI tools can be used by IT staff for building solutions for end-users; there by increasing productivity.

Might 2

Variety of AI application used different levels for various purposes.

Magnified intelligence

Building systems by extending program intelligence with capabilities like prediction, classification, rule-based or conditional reasoning using classical AI algorithm like Vector Machines, Naive Bayes and Random Forest.

Towards predictability & precision

Complex systems with greater generalisation, increased accuracy, and capable of learning, etc., for example neural network-based machine translations and classical insights implemented using deep learning algorithm.

Generative or creating

Human like cognitive content creation using distributed learning. Video insights, music creation, etc are some examples where new content is created using trained data.

Summary:

Generative AI is just another type of AI that can help humans in creative work. Most of the Gen-AI works are closely associated with human interaction and support. Even though AI is being researched for decades, the GenAI is most adopted.

Just started, let’s discover more on ‘Generative AI’ in following blogs.

Glossary:

• Myth(noun) — mostly believed information but different from fact.
• Might(noun) — the power, energy, or intensity of which one is capable of.