The Evolution of AI: Traditional AI vs. Generative AI

AI powered tools have been used for decades, but the recent breakthroughs in generative ai have pushed the topic front and center, but what's really different about new generative ai models like large language models (LLM) compared to the traditional AI. In this video Shad Griffin explains the fundamental difference in their architectures that have made Generative AI possible.

[christopherpetersen]

sort of an odd description that didn't mention symbolic logic, rule-based systems, optimization problems, etc. historically. the feedback loop is about machine learning in the context of self-improving systems and doesn't necessarily imply artificial intelligence at all. and, it's not about holding lots of data lately, it's about unsupervised (usually) learning over massive data sets that aren't exactly stored in the model or the application layers you drew. no mention of transformers or foundational models?

[Dr.SusantaMitra-gxhy]

A useful video with nice explanations. Keep up the good work. I also create videos on AI-related technologies. Thank you.

[Stan_]

How do you do the drawing ? Is it from behind glass or some other nice trick ?

[MrDrivingFaster]

In comparison to Predictive AI, Generative AI appears more impressive because of the LLM and the broader data set, that optimise the feedback loop. In other words, Gen AI predicts a better answer and formulates it in a construct that is more to what a human would expect.

[frankgreco]

This is a misleading title. It should be "Traditional Business Analytics vs Generative AI". It subtly puts Predictive AI in the camp of "traditional", as in old or dated, which is hardly the case. The economics of Predictive AI far outweigh Generative AI currently (see Andrew NG). Certainly, GenAI has a bigger growth curve, but PredAI is still the right tool for many use cases. If you want to talk about traditional AI, you should talk about the 80s boom/bust of Prolog, Symbolics, LISP, expert systems, et al.

[michaelpaczynski]

Thank you, but I’m not convinced. The fundamental architecture from their early days of the Turing machine still apply. The difference is where the repository is and the sophistication of the mathematics.

[danzinde]

Although a simplistic explanation, this video missed the point it wanted to convey.

Traditional AI is not only about data and analytics. You can talk about rule- based NLP kind of systems used in expert systems modeling and knowledge representation. And add Machine learning to process massive data and come up with predictions that AI could use as recommendations or taking appropriate actions.

GenAI can still work on internally trained data. In fact, the idea is to train LLM on customized datasets to build business specific systems. The concept of the foundation model should have been mentioned here to link traditional AI on how foundation models are built using established machine learning techniques like deep neural networks.

[maikvanrossum]

So basically ML models vs. LLM…? Because ‘traditional’ makes it sound like ‘classic’ or ‘vintage’ vs. ‘modern(ized)’ and ‘only new’. There are scenarios and use cases where the LLM runs inside or on-prem, or the dataset is scoped and therefore reduced. So yes, I can see the differences between the two architectures shown here, but this comparison I think is confusing.

[rcstscc]

Can we evaluate generative AI as "AI as a service" in this framework?

[YesCivic-R]

Most human being make the same mistakes, many generations. That Feedback loops can predict pretty much the same behavior generally, then deep learning helps a lot to be more accurate with require scale and HPC.

[amritbro]

Generative AI have loads and loads of data which is the prime reason for it for being so much powerful than that of traditional AI.

[BillAnt]

If you think about it, even early videos games like Pong and Pac-Man has a primitive "AI" since it had to know or predict your moves when playing against the machine. A better example is computer chess games with complicated strategies.

[rdhanur]

Great Video. Could you please share some Oil & Gas Subsurface project that using Traditional AI and another project using Generative AI?

[IcyAmphibian]

JASMY, built on IBM’s Hyperledger Fabric, combines generative AI, blockchain, IoT edge-network, and decentralized GPU services for AI computation. Imagine all of that Earth data in a secured and distributed IPFS, with NFT technology tying the data locker IPFS address to the user in an anonymous and private manner.

[molugusatyapriya]

Great breakdown of how traditional AI stacks up against generative AI. This video really highlights the transformative potential of the newer technology

[Lineberek]

You could use some practical examples to help those who are not very familiar (like me).
I assumed the first part was about Excel.

[Jagentic]

A naïve question: with advances in quantum - and the specific q’s they’ve been able to demo astounding results - how far are we from perhaps an AI training run that could be formulated in such a way as to use quantum’s enormous efficiency and capacity - even if only for a specifically structured single-ish shot? Still a longshot?

[AatishN-ju]

Beautifully explained. Though it took into consideration a simpler use case of AI, it certainly break it down very well.

[kharesiddharth]

Difference is the gen ai generates something like an image or poetry vs ML /ai is more for classification and labeling . And yes gen ai is usually using large language models / nural network

[paolo.miscia]

I found this very useful. Might be a bit... simplistic, but it really helps differentiating the two.