From Generative AI to AI Agents, how AI is changing the landscape.

Over the last two decades a lot has changed in human computer interaction, the initial version where humans interacted with computer terminals required programming expertise to the latest version where you can talk directly to a computer application. In this article we will go over the evolution of human computer interaction and how generative AI and AI agents have evolved over time. Let’s get started with this illustration:

Computer Terminal
Hi -> command not found

LLM (GPT/Gemini/Llama)
Hi -> Hi, how can I assist you today?

AI Agent
User: Hi 

AI Agent: Hi, it’s Friday afternoon. Do you fancy any delivery for dinner?
User: Yes 

AI Agent: Last time, you ordered food from a 4-star restaurant for 2 people. Would you like to order something new?
User: Yes, can you order for 6 people this time?

AI Agent: Awesome! Thai food for 6 people will arrive by 7 pm. Enjoy.

We’ve come a long way from computer terminals that didn’t understand human language to smart voice assistants that can set timers and adjust thermostats. Now, we have conversational LLMs and AI agents that can think and perform tasks based on goals to assist us.

What Alexa and Siri Couldn’t Do

Alexa, Siri, and Google Home (and others) are first-generation smart voice assistants. They were trained with variations of inputs and possible answers. For example, if you ask, "What’s the weather today?" the response would be something like, "The weather today is X degrees with a high of Y and a low of Z." They perform well on questions they’ve been trained on, but they have their limits.

Enter GPT/BERT:

With models like GPT and BERT, the approach is more brute-force: feed them any text, not just Q&A like Alexa and Siri, and they learn patterns from billions and trillions of tokens. A token is roughly 3/4 of a word. Once trained, these LLMs can generate human-like responses in chat mode. While they might sometimes produce factually incorrect responses (a phenomenon known as LLM hallucination), the answers look realistic and almost correct.

The Revolution in Generative AI Technologies

The advent of advanced conversational AI capabilities, sentence completion, and sentiment analysis has been a game-changer in the tech industry. This breakthrough was not limited to a single organization, but rather a collective achievement of the AI community.

The real disruption came with the democratization of this technology. When it was made accessible via APIs and user-friendly chat interfaces, it quickly scaled to hundreds of millions of users worldwide. This led to a scramble among tech giants and gave rise to a new wave of products, such as Mistral, positioning itself as a European response to this technological advancement.

The power of generative AI has been recognized globally, and its impact is evident in the tweets and discussions among tech leaders. From debates on the partnership between tech giants and AI organizations to the launch of AI models by Chinese tech companies, the influence of generative AI is undeniable.

However, the journey towards perfecting AI is still ongoing. As we continue to train AI with trillions of parameters and unbiased data, we strive to fill the knowledge gaps and work towards creating a more personalized, user-centric digital world.

The future of generative AI is promising, and the next wave of innovation is just around the corner. Let's continue to push the boundaries and shape the future of AI together.

Soopra AI: Democratizing Personalized Learning with AI

Soopra AI, founded by Praveena Dhanalakota, is a fantastic example of how advancements in Generative AI are driving innovation. Similar to Khanmigo, an AI-powered tutoring tool by Khan Academy, Soopra AI offers AI personas and interactive courses created by top experts. 

While Khanmigo focuses on helping students with various subjects, Soopra AI takes it further by allowing users to connect with AI personas of real experts. Imagine learning directly from AI versions of top professors and industry leaders! Soopra AI’s courses feature daily 10-20 minute lessons with 24/7 chat assistance, designed by teachers from institutions like Stanford and Harvard. This ensures you get the most current and relevant information.

Soopra AI’s personalized learning paths and real-time support make it a more comprehensive platform compared to Khanmigo. Whether you’re looking to dive deep into a subject or get practical industry insights, Soopra AI has got you covered. It’s like having a personal teacher and a mentor rolled into one, making learning not just effective but also super engaging.

Wrap Up

There are tons of interesting topics in Gen AI and LLM, but today’s post highlights how access to APIs and apps drove exponential adoption and disruption. In the next post, we’ll dive into RAG and Agent Architecture. What other AI/Gen AI topics are you curious about? And if you’re interested in learning more about AI and engaging with expert AI personas, check out Soopra AI now.