Pular para o playerIr para o conteúdo principalPular para o rodapé
  • 12/07/2025
What is AI and Where is it Today - Explained in under 9 minutes!
Transcrição
00:00What is AI and where is it today?
00:03Explained in under 9 minutes.
00:06This is a follow up to a video I created about 7 years back titled, What is AI in 5 minutes?
00:13Since then, a lot has changed, so here is a sequel with additional information.
00:18The information in the original video is still true, so think of this as part 2.
00:25When we last talked about AI, it was getting good at recognizing images, understanding
00:30speech and analyzing data.
00:32Today, AI is creating, reasoning, making decisions and even interacting like a human.
00:41Some of the biggest breakthroughs are in Generative AI, Agentic AI, Autonomous AI and the journey
00:48towards AGI or Artificial General Intelligence.
00:55Let's first start with Generative AI or ChatGPT kind of models that most of us are familiar
01:00with by now.
01:03Gen AI has gone beyond just analyzing data to producing new content from scratch.
01:10Gen AI is built on large language models, like GPT, to predict words by analyzing massive amounts
01:18of text and learning their probabilities of occurring close together.
01:24They use deep neural networks, specifically something called transformer architectures, which are
01:30a type of neural network designed to handle sequential data efficiently.
01:37Transformers improve upon traditional neural networks by using self-attention mechanisms, allowing
01:44the model to weigh the importance of different words in a sentence and understand the context
01:50over large or long passages of text.
01:55Diffusion models create images by starting with some random noise and refining it step-by-step.
02:02They work by gradually reversing a process that initially adds noise to an image, training the model
02:09to reconstruct the original image from the noisy version.
02:14This is achieved through deep learning networks, typically using something called U-Net architectures,
02:22which learn to predict and remove noise at each step, ultimately generating high-quality realistic
02:29images.
02:30Multi-modal AI integrates text, images, and audio, allowing AI to understand and generate
02:39content in multiple formats.
02:42It achieves this by using specialized transformer-based models called multi-modal transformer model that
02:50can process and align different types of data within a shared representation space.
02:57As you know from my previous machine learning videos, AI's main way to understand the world
03:03is to map concepts in some representation space.
03:07This is often a vector space.
03:10In here, by learning relationships between the modalities, multi-modal AI can perform tasks
03:17like describing images in text, generating images from text prompts, or even creating videos
03:24with synchronized audio and captions.
03:29And then there is agentic AI.
03:32Traditional AI follows instructions, but agentic AI makes decisions and executes tasks independently.
03:40It uses memory, planning, and reinforcement learning to execute tasks without human intervention.
03:50It does so by breaking down complex goals into smaller steps, adapting when things go wrong, and
03:57continuously improving.
03:59One of the common use cases for agentic AI is that it integrates LLMs with external tools
04:06like databases, APIs, and reasoning frameworks.
04:11This allows AI agents to go beyond simple text generation and interact dynamically with external
04:18systems, retrieving real-time information, performing calculations, and executing automated tasks.
04:26For example, when a customer asks, where's my order?
04:32Instead of just providing a generic text-generated response, the AI can query the order database
04:39to retrieve real-time status, call the shipping API to get the latest tracking update, analyze
04:46previous customer interactions to offer proactive support like offering a discount if the order
04:53is delayed, and automatically send an update to the customer, reducing the need for human intervention.
05:01Agentic AI relies on deep learning for perception and language understanding, but also integrates
05:08with symbolic reasoning, memory, and goal-oriented planning, which go beyond deep learning.
05:18The AI we just discussed was all digital, but now AI can interact in the physical world as well.
05:26For example, robotic AI combines sensors, reinforcement learning, and computer vision to navigate real-world
05:35environments.
05:36We know about self-driving cars navigating the roads.
05:41They use sensor fusion, which is a combination perhaps of LiDAR, radar, and cameras, plus neural
05:49networks to detect objects and predict motion.
05:53Technology-wise, this type of AI uses deep learning, control systems, reinforcement learning, and
06:01symbolic reasoning.
06:04And then there's the end game, which is artificial general intelligence, or AGI.
06:10Right now, AI is still narrow.
06:13It can master one task at a time.
06:15The dream of AGI is where AI can think, learn, and reason like a human.
06:23How would that work?
06:24Instead of just predicting outcomes, it would use causal reasoning to understand why things
06:29happen.
06:31It would use few-shot learning to learn from a few examples, just like humans, unlike today's
06:38AI, which requires a huge data set.
06:42Here are some optimistic examples and predictions of how the future would be shaped by AI.
06:50Personalized assistance AI assistants are evolving to predict what you need before you even ask,
06:56from scheduling meetings to reminding you about daily tasks.
07:00Over time, they learn your habits, preferences, and routines, making life smoother and more
07:06efficient.
07:07AI-Human Collaboration AI acts as a powerful tool that helps humans
07:14think faster and work smarter, whether in business, art, or science.
07:19Instead of replacing jobs, it automates repetitive tasks and provides insights, allowing people to
07:26focus on creativity and complex decision-making.
07:31AI for science and medicine.
07:34Instead of slow trial and error testing, AI can analyze millions of potential drug compounds
07:40in a fraction of the time.
07:42This accelerates the discovery of new treatments, helping researchers find cures for diseases
07:47faster and more efficiently.
07:51With all this going on, how can you stay ahead?
07:54The best thing you can do is to continue learning how AI works, experiment with tools, and stay
08:01informed.
08:02AI can help you work smarter.
08:05AI literacy is as important as digital literacy.
08:10Understanding AI will be a key skill for the future.
08:14If you enjoyed watching this video, please like, share, and subscribe.
08:18For a one-page visual of this and all future videos, please sign up on my website.

Recomendado