AI Revolution
One can argue that in 2026 AI is revolutionizing the Global economy. AI is everywhere and many if not most companies in the world have plans in place for AI usage.
I like to think AI is both revolution and evolution as in the next phase of advancements in technology. We can go back to Alan Turning’s theories in which a machine can be programmed to perform any task by giving it a set of logical instructions.
Artificial Intelligence or AI is the advancement of information technology which allows computers to perform tasks that would normally require human intelligence.
Basically, AI learns from the consumption of data and identifies patterns in the data which it can use to make decisions and solve problems efficiently.
A subset of AI is Machine Learning where machines learn by analysing patterns in data collected.
Deep Learning leverages Neural Networks to handle complex tasks like speech.
The Generative AI part of deep learning creates content via text, images or music.
Large Language Models (LLMs)
By definition a Large Language Model or LLM is a type of AI model trained on an enormous set of accessible data, think everything published on the internet. It understands human language as well as code, music, videos, etc. It uses probability to predict what comes next based on the huge consumption of data it has done.
Hallucination occurs when the prediction turns out wrong. This is because AI results are relayed being most likely and not absolutely factual. Sometimes these predictions can be nonsensical where they appear to be like hallucinations.
Responsible AI: Ethics & Limitations
Despite its power, AI has significant hurdles that require human "grounding":
Knowledge Cutoff: AI is limited by the date its training data ends.
Bias & Harm: AI can reinforce societal biases (Representation Harm) or perform poorly for specific groups (Quality of Service Harm).
LLMs can auto-complete sentences using next-word prediction based upon probabilities.
Embeddings will convert words into lists of numbers which allows the computer to understand how words are mathematically related.
Transformers and Attention allows for the model to consume the entire phrase and determine the context of the phrase as a whole instead of a sequential process of individual words.
Google AI Evolution
Google has and still is a major contributor to the AI Revolution.
It started using Machine Learning in search levering spell check and suggested words. Then Google translate started to use Statistical Machine translation leveraging enormous amounts of data to find patterns that translate text.
Machine learning API has been made available for use leveraging REST API’s.
Google Cloud Translation API allows for language translation.
Cloud Vision API provides the ability to analyze, identify and extract information from images.
Cloud Speech-to-Text & Text-to-Speech converted auto files to text and test to audio files.
Cloud Natural Language API introduced sentiment analysis and entity recognition which provides the understanding, emotion and entities from text.
BERT. Using the Transformer, Google released BERT, a model that could understand the context of words in a sentence by looking at the words before and after them. This was the biggest leap in Search history.
LaMDA. Announced at Google I/O, this was a breakthrough "Language Model for Dialogue Applications" designed to have fluid, natural conversations.
PaLM (Pathways Language Model). A massive model with 540 billion parameters that showed "chain-of-thought" reasoning, allowing it to explain its logic step-by-step.
2023: Bard. Google’s first major conversational AI was launched, originally powered by a lightweight version of LaMDA.
2023: Vertex AI & Model Garden. Google Cloud expanded its tools for developers, creating a "Model Garden" where companies could choose from various AI models (like Vision, Speech, or Text) to build their own apps.
2024: The Gemini Era. Google rebranded Bard and its other AI efforts under the Gemini name. Gemini was built from the ground up to be multimodal, meaning it can "see," "hear," and "read" simultaneously.
AutoML: Allows users to create custom models with minimal coding (e.g., training an AI to recognize specific parts for your business).
Vertex AI: A professional "full-stack" platform for building, scaling, and managing the entire lifecycle of custom neural networks.
Infrastructure: Raw power using GPUs and TPUs (Tensor Processing Units) for massive computational tasks.
AI is viewed as a "people-first" tool designed for Augmentation (improving human-led tasks) and Automation (handling repetitive work).
Implementation Steps:
Define & Gather: Identify objectives and prepare the "lifeblood" of ML—data.
AI in Data Analysis
The 5 Steps to Implementation
Define your objectives.
Gather and prepare your data.
Choose the right tools.
Start small and experiment.
Learn and iterate.
AI can provide insights, create datasets for training, predict market outcomes, forecast demand, and monitor business performance.
AI can analyze Structured Datasets (spreadsheets), Unstructured Datasets (text, images, video), Time-Series Data (stock prices, weather), and Big Data.
Gemini
Google Gemini is the latest and greatest Large Language Model form Google. A Multimodal model it handles text, video, images, sound and code as both input and output. Also, with permission can connect to personal email, docs and photos. Agents can execute tasks like scheduling meeting, creating presentations, accessing databases, etc.
Let’s start with the + icon
The + icon lets the user add content for model input in addition to the text you can enter in the input box.
Upload Files - Allows you to browse machine accessible directories like an attachment in an email. Files can be images, documents, videos, basically any file can be uploaded for model processing. Once uploaded you can instruct the model to summarize text, analyze data or extract content from files.
Add from drive - Provides a way to add input for model processing via Google drive.
Create image - Quick way to create image by providing format and picture.
Create Music - Same as create video with music
Next let’s focus on the Fast drop down at the end of the input box
Fast indicates the model you use when processing the input. By default it uses the Free tier Gemini 3 Flash model for instant responses.
Thinking should be selected for complex queries like system design, multi step problem solving or in depth document processing. It will provide results slower than fast.
Pro selects the Gemini 3.1 Pro model. It uses Advanced Reasoning for r complex problem-solving>
No comments:
Post a Comment