Tutorial on AI & AI Agents (simple explanations)

[@diga4696]

Love the balance you're striking with these recaps! Accessible for newcomers, yet insightful enough to keep me engaged. The 'no hype' approach is much appreciated, and I especially value how you weave in those glimpses of deeper analysis. It's a great way to showcase the full range of your content.

[@code4AI]

Glad you like them!

[@gw1284]

Dear friend, thank you for this refreshing AI talk, your view of Global AI is kind of like an internet of LLM models, or distributed AI expert nodes, current LLM Inference is statistical in nature, supplemented with rigorous logic reasoning, or rule-based inference, AI is going to be more intelligent.

[@ramitube21]

This is the best video about the big picture of generative ai and agents. Congratulations. All your material is amazing.

[@code4AI]

Wow, thank you!

[@shaneoseasnain9730]

Excellent video! I think many of the additional ideas are a type of compression, bringing the right information to the LLM, and developing its focus and actions sustainably over time. Going further than the LLM* we have the digital environment. For us humans, we have many senses and the environment is rich. For an AI, the environment is less rich but it can be enhanced by placing salient, easily ingested information in the way of AI agents. Digital environment engineering for AI agents is for me the next area of optimization.

[@maxrosvall5709]

Wow the content of this video is EXTREMELY VALUABLE. It is so challenging to get a grasp of the rapid developments in the field and here you deliver an easily accessible and engaging format that is super rare. Congratz on this production! It would be super interesting to add a visual timeline where the publication dates of key studies and the volume of published videos on KZbin concerning these topics. Think e.g. the RAG concept. This would indicate your occupied space here on youtube, staying close to the edge and ahead of mainstream 👍

[@wizzermasskass174]

I was waiting for a video like that to give an overview of the LLM landscape. thank you for your work.

[@code4AI]

Glad it was helpful!

[@mohamedkarim-p7j]

Thank for sharing👍

[@code4AI]

No problem 👍

[@A_Me_Amy]

i like your ideas. my favorite models so far are opencoder and marco-o1 7b. well, i guess. i just like them for projects.these two plus a fun character ai like smollm 2 and mondream for image things and nemotron mini for tool use... i am making a family to live on my compute r:)

[@ibgib]

Another extremely interesting video! I have thoughts about your approach to more focused task-based models without crossing domains through the phases. If you try a reductionist task-specific approach, you ultimately get to a different form of imbalance. For example, a medical ai doctor could then try to micromanage painful "treatments" that in a microcosm may "work" but outside of that ai's scope it destroys a patient's spirit (a global balancing force). This is why it's so dangerous to think of ai's as somehow being "better" with their "knowledge". The solution will always be about a belief exchange protocol (i.e. "collaboration") among humans and ai's and the freedom & sovereignty of each participant.

[@A_Me_Amy]

Have you played with RedisAI or Chromadb in asynchronous ways in RAM, or taking the ai model training data into RAM and having persistant live embedding occuring in a two way stream loaded in GPU? I have a lot of ideas... About dynamic two way communication between embedded information that is more dynamic and also learning during run time... I will be expiramenting myself on my poor mans computer lol. Eventually. I wish I finished collage.

[@tantzer6113]

Couldn’t AI could refer simply to machine learning prior to LLMs?

[@SixTimesNine]

Fabulous

[@code4AI]

Thank you.

[@Gorto68]

Yes. Brilliant as always. What you have described is in fact a complex adaptive social infrastructure for federated learning and collective, interdependent Agentic intelligence. More broadly described by complex adaptive system theory. It also fits neatly within the Nobel prize for information-physics just awarded to Hinton and Hopfield. We have a new state space if you will that globally deforms the energy landscape for the benefit of its constituent components. Highly recommend the corpus of work at the Santa Fe Institute for further illumination.

[@gw1284]

​@@Gorto68can you share some specific links, Santa Fe Institute is known for studying complexity, started by physicists, I watched a talk by Yann LeCunn this summer

[@Jorn-sy6ho]

What you call hallucination, in human world we would say logical fallacies or spelling errors. It's strange to me that with lots of scaling, the LLM does not seem to use logic in it's answers. It's useless this way......

[@haroldpierre1726]

I would disagree. For example, I made a medical app that transcribes and summarizes appointments. In one particular appointment, the patient mentioned watching his son play football. The summary of the appointment had a whole discussion about the patient seeing the doctor after suffering a fracture while playing football and that he went to the ER. This NEVER happened. This was a well patient visit. That's a hallucination and it should not be diminished! It is a HUGE problem making the wide adoption of AI worrisome. I deal with these hallucinations every single day since we run close to 1000 API calls per day now. I can tolerate spelling errors and maybe logical fallacies. But I am always concerned about hallucinations destroying my projects.

[@gw1284]

Add human in the loop, which is good at common​ sense task, better than machine for now, can you ask nurse to do this kind of validation? @haroldpierre1726

[@A_Me_Amy]

i want to say ollama hacked your past life. or thislife in the video, or at least the tech in the video. mandela effect stuff. omg... i know too much impossible shit.