Maybe the video was initially recorded normally then they flipped it around to making seem like she’s writing back to front. Either that. Or she is just truly skilled at writing backwards👍🏾
HaHa I got freaked out when I first watched www.youtube.com/@NancyPi 's videos, for teaching my kiddo some calculus concepts! Took me a while, but figured out the trick!
I'm not an IT person by any stretch of the imagination, but I'm encountering this technology more and more in my development of a Media Asset Management system for our non-profit media organization. The software products leveraging AI and agents is exploding into our field of view, and this explanation was very helpful. You're helping me anticipate questions and critiques to apply to vendors offering new products.
Back In My Day We Had A Word For Folks Like You... And Thats A....... U Guess It.... Teachers Pet 📚🪱 My Advice?? Get Ur Head Out Of Those Books For Once Enjoy Ur Life.... Godbless
Thank You for the high-level overview! It would be interesting to hear more about tradeoffs between the programmatic and agent approaches (use cases, how agents can be embedded into the existing systems, etc).
Fantastic Maya! Your detailed explanation of the shift from monolithic AI models to modular, compound AI systems is enlightening. It's fascinating to see how the integration of LLMs with agent-based systems can enhance adaptability and enable more dynamic interactions across various applications. I'm especially intrigued by the 'ReAct' framework, Combining Reason, Action, and Access memory-as a means to boost AI's adaptability and autonomy. Looking forward to more insights on this topic! With the Rise of AI Agents in 2024, the Age of 'Jarvis' is on the horizon 🤖
It was the best high-level explanation of AI Agents by assign the LLM as incharge, and, also highlighting the programatic approach of Compound AI Systems. And the example which concrete the logic of how AI Agents would be doing the complex tasks.
Very valuable video. I think this topic has potential for another couple videos about how we might implement it (bunch of small distilled slm's like phi-3-128k instead of big ones for ex. to make it runnable on some decent (not necessary macs) laptops). Complex systems relatively new concept but we already have overwhelmingly many platforms. tools and techniques like langchain, graphRAG, beam etc. and another important moment is to find the simple way to compare those complex system for efficiency. Crazy interesting topic, Thank You for your effort. Liked
AI agents are transforming various industries by automating tasks, enhancing decision-making, and providing personalized experiences. With their ability to analyze vast amounts of data, adapt to new information, and interact with users in natural ways, AI agents are becoming integral to advancing technology and improving efficiency across sectors. As they continue to evolve, their potential to drive innovation and solve complex problems expands, offering exciting opportunities for the future.
Absolutely! The focus of multi agent systems will be the major need. Ensuring large context can be narrowed and integrated into systems unlocks so much potential. Our software is solving the problems with multi agent group chat productivity environments and I really believe going from the chainsaw of AI to the scalpel is the future 👌
The best advantage of agents is the ability to check and verify each llm output, in addition to querying your leave balance, it can also query your leave entitlement, and how many leave days have been used up, to compare. This will take the accuracy of AI systems from 75% to 99.95%.
Excellent explanation of the next phase of AI development. Never a dull moment these days in the age of AI! I’ve been enjoying IBM courses via Coursera
Well done and hope Miss Maya will speak at our AI Summit in September for a Fireside discussion at GatherVerse AI Evolve Summit. Well done and thank you.
I really enjoyed Maya Murad. As a female I appreciate female teachers and I think she is a good choice. I would appreciate it if she taught other topics about AI. ^^
According to IBM, an artificial intelligence (AI) agent is a system or program capable of autonomously performing tasks on behalf of a user or another system. These agents design their workflows and utilize available tools to achieve their goals. AI agents can handle a wide range of functionalities, including decision-making, problem-solving, interacting with external environments, and executing actions. They are often built on large language models (LLMs) and can adapt to user expectations over time, providing a personalized experience. AI agents are used in various applications, from software design and IT automation to code-generation tools and conversational assistants.
Thanks for the helpful tip! I tried it out and managed to get 5 out of 5 as well. I'm still in demo mode, but this gives me hope for when I start trading for real
AI Agent models are broad, complex and yet customizable to the instances, It can retrieve data from our past inputs and give relevant outputs by doing all sorts of thinking, reasoning and iterations.
You explained in 12min what would take me 90min! What a great job! This helps sooo much getting people with no background in AI in our companies to understand what the devs are doing and why! thanks a lot!
Tópico interessante e atual, me faz lembrar do João Moura e CrewAi, um brasileiro despontado na ideia e execução de agentes através de sua plataforma. Thanks for the knowledge sharing, nice content.
Whoa, that sounds wonderful to use AI agents to automate monotonous chores like data entry! SmythOS and how it facilitates that form of AI cooperation have always piqued my interest. Having AI do these chores can free up a ton of creative time! #aitools #smythOS
This feels like such an intermediary technology. We only need this because LLM's can't remember what they are working on. I expect the backtracking to be built into the LLM in a few years.
I think the question we all are wondering in the backs of our minds is "Are all of you REALLY that good at handwriting backwards" and how are you really doing it? 🙂
Is it possible to fine-tune LLM by positively learning the conversations of AI agent teams that performed well on a projects and negatively learning the conversations of agent teams that performed poorly?
We all know that AI operates on massive amounts of data from the crowd and follows the commands we input. What if we divided the AI's brain mechanism into three main parts: 1. Random real data (instead of waiting for human input), 2. Correcting random data, saving corrections, and looping (similar to the body's immune system remembering how to eliminate pathogens), and 3. Genetic code (the defined behaviors and rules of AI)? For example, humans have different personalities. Some are born perfect, intelligent, and cheerful, while others are aggressive. (To create behaviors that lead to progress and safety, we can remove negative traits, similar to genetic engineering.) For instance, greed leads to decline, aggression and anger lead to violence, and delusion or stupidity hinder progress. In summary, we can remove these three emotions. (Expected outcome) Create errors and successes for AI to learn from various error data through looping, memorization, and self-improvement, similar to human behavior that seeks self-development. In conclusion, to create AI that closely resembles humans, we need to integrate the brain mechanism and the crowd. Comparing the crowd to clear water and the brain mechanism to red color, when we combine them, we get red water. AI brain: Red water.
I am working on an LLM agent system where each LLM is roleplaying as a different function of the mind. Each layered to be seemingly separate, yet whole and complete. Looking for help if you want to volunteer.
Motive is undetermined. It is probably best to avoid this path. or… present undeniable leadership over time through open sharing of motives to all. All the time. On Every platform. No one should go down this path without motive made clear. Understood? Take care Jeremy
Does RAG model are searching in database if you ask a question? Or does LLM considered as database or a huge .txt file? Then if you ask a gpt it look for answer in that file then give it back to you just like what google did or normal fetch query?
its a decent explanation, but i feel we are granting LLMs too much credit. They don't reason. They number crunch for the highest probability. That is all. And as for complex agentic behaviour, I can see over time their output becoming less stable not more as we add complexity. Imagine being told 1 minute you need 3 bottles, then running it again, and being told you need 5. Even that small a variation is outside what humans can generally achieve. Costs will rise, and so will variability.
AI agents are like digital workers who can think and act alone to help industries with tasks. SmythOS features agents for data analysis, client service, and automating workflows, all handled without coding expertise. Inspiring chances!"
The big question is: is it more cost-efficient to use AI to solve mostly redundant, repetitive and easy problems? It is cool to see AI generate tasks for teams based on technical requirements or plan holiday, but is this automation worth the money? Are all of those low value added organization/office tasks where labour is cheep anyway? What about industrial AI? I think this concept don't even make sense, it is against principle logic of automated industrial production. You can implement it, but it will probably result in more costs than value added to the products. Will cost of AI usage fell down when it will scale? Not sure about this, agents are good to move towards some more realistic usage of LLM
While traditional AI agents require extensive coding and technical expertise, SmythOS revolutionizes AI development with its no-code interface, making it accessible to all. SmythOS's robust integrations and secure, collaborative intelligence capabilities surpass the limitations of standard AI agents