![๐Ÿ”ฎ Artificial Intelligence yesterday, today and tomorrow](./slides-export/1.png) I'll speak fast, let's chat together about questions during the breaks? ![Machine Learning (ML)](./slides-export/10.png) The idea of ML is ~ just to treat computers as ๐Ÿ‘ถ babies, instead of _programming_ them! For example: ... LLMs basically do the same to be able to reply to prompts where you chat with them and ask them questions. The basic idea is not that new ([backpropagation](https://en.wikipedia.org/wiki/Backpropagation) ~1980s?)... ...but only recently (~2010s+ ?) it suddenly turned out to be a lot more ๐Ÿš€ interesting, due to the emerging increasing availability of _Big Data_ and massive storage & _Super Computer_ infrastructures in _โ˜๏ธ Clouds._ ![Magic?](./slides-export/11.png) Is ML ๐Ÿช„ magic? Not at all... the basic idea is really quite simply, actually! To illustrate: ![undefined](./slides-export/12.png) Picture says more than words? But how did we find the a & b parameter... ![undefined](./slides-export/13.png) We literally just make a computer program try out values for a and b, to try to make the model have "good accuracy" - in this case, that just means "making the red line as close to those green points as it can be". E.g. a large language model really is (kind of) similar to this - except that instead of have 2 parameters, for a and b, it has more - many more... Copyright ยฉ 2020-2021 Gajanan Bhat. All rights reserved. ![undefined](./slides-export/14.png) ... imagine, that instead of only having to find x2 parameters, a & b, for X car weight and Y gas (or ๐Ÿ”‹ electricity!) consumption, you need more parameters, e.g. for Pie Sales, Price, and Advertising cost? PS: This is technically mathematically not entirely accurate (because it's not really about spatial dimensions, but function parameters; really a quadratic or N polynomial function on 2-D X & Y would in a sense be more accurately making this point) - but this is a just a better visual illustration! ;-) ![๐Ÿง  Your Brain is a Biological Neural Network](./slides-export/15.png) If in addition to remembering Linear Regression from your High School Math class you also remember a little bit of your biology 101, then perhaps this image is familiar to you? It's ![undefined](./slides-export/16.png) ![GenAI ML Models](./slides-export/17.png) Instead of just 2 or 3 such parameters, as seen previously. In reality often not just 1 model, but Pipelines with Workflows connecting several models; e.g. LangChain's LangGraph in FLOSS. ![Training & Inferring](./slides-export/18.png) Pictures of digits, cats & dogs, or texts... ๐Ÿ“ฑ Mobile Phones can โœจ do (~) inference locally. No โ˜๏ธ Cloud (DC) is more ๐Ÿ” #private & #fast. ![Data?](./slides-export/19.png) ![๐Ÿซข](./slides-export/2.png) How the heck does this work?! And what does all this mean for the future? BTW, you can try this out for yourself on gemini.google.com ... Prompt: _Can you make this sound cooler?_ _Michael Vorburger has been fascinated by the creative power of computers from an early age. He is an avid hacker by night, see https://www.vorburger.ch, and currently works on ML technical infrastructure in management for Google by day. He is presenting at this event in his personal capacity, not representing his employer._ ::right:: # Output ๐Ÿซข How does this work? What does this mean for the future? _By night, Michael Vorburger transforms into a digital alchemist, coaxing unseen potential from the very ones and zeros that form our world. An avid hacker with a website that whispers secrets of the craft (https://www.vorburger.ch), he spends his days architecting the intricate machinery of Machine Learning infrastructure for Google. While tonight, he sheds his corporate cloak to share his personal visions, untethered and electrifying._ ![Energy?](./slides-export/20.png) https://www.tomshardware.com/tech-industry/artificial-intelligence/using-gpt-4-to-generate-100-words-consumes-up-to-3-bottles-of-water-ai-data-centers-also-raise-power-and-water-bills-for-nearby-residents ![undefined](./slides-export/21.png) ![Cloud? Open Source?](./slides-export/22.png) ![undefined](./slides-export/23.png) ![undefined](./slides-export/24.png) ![Personalized LLM](./slides-export/25.png) ![E.g. Gemini Extensions](./slides-export/26.png) ![undefined](./slides-export/27.png) Scotty tries to talk into the mouse of a computer in Star Trek... this used to be a joke, but... we're there now. ![Gaps?](./slides-export/28.png) Combining Large Language Models (LLMs) with Knowledge Graphs (KGs) and Retrieval-Augmented Generation (RAG) looks promising. You would be surprised how difficult it is to build an Assistant that can buy movie tickets or make restaurant reservations only. Something like Gemini's Extensions, to connect it to your Gmail & Google Docs, Google Maps or Flights, or YouTube are only the very early beginning. Google Search Generative Experience (SGE) ![undefined](./slides-export/29.png) I'm not a cars sort of guy, but if I were, I would be pretty excited to get my Knight Rider KITT, any day now... ![undefined](./slides-export/3.png) ![The Future?](./slides-export/30.png) I didn't come here with a horse carriage, like I would have just a 100 year ago; and we don't have coachmen anymore, we didn't need that job anymore. But in the bright side, there A LOT of new jobs nowadays which didn't exist 100 years ago. Turing Test, Robot College Student Test, Employment Test, Ikea Test, Coffee Test, Modern Turing Test ![Applications](./slides-export/31.png) AlphaProof & AlphaGeometry solved 4 / 6 problems from this yearโ€™s International Mathematical Olympiad (IMO); https://www.theguardian.com/technology/article/2024/jul/25/google-deepmind-takes-step-closer-to-cracking-top-level-maths Khan also e.g. https://www.youtube.com/watch?v=_EfEoSP7oYQ (after aforementioned TED Talk) TODO Try Google Lens with Homework & screenshot it https://c2pa.org for GenAI? ![undefined](./slides-export/32.png) Google DeepMind's AlphaFold "protein folding" breakthrough unlocking research of new medicines ![undefined](./slides-export/33.png) To quote Dr. Phlox, from Star Trek: "Optimism, Captain!" # Optimism < ! -- Anything that can go in an HTML video element. - - >

Your browser does not support videos. You may download it here.

![You?](./slides-export/34.png) ![undefined](./slides-export/35.png) https://en.wikipedia.org/wiki/Artificial_intelligence TODO Expand this, modeling it with (my) https://docs.enola.dev, similar to (TBD) https://github.com/enola-dev/enola/blob/main/models/enola.dev/other.ttl ![undefined](./slides-export/4.png) https://www.jasondavies.com/wordcloud/ ![History ๐Ÿฏ](./slides-export/5.png) AI arguably started in antiquity - the 1st link is to an interesting "AI history" sort of table on Wikipedia, which mentions e.g. ancient Greek myth of Talos, the giant automaton in Crete from ca. 700 BC, or the jewish Golem, or the Takwin of Muslim alchemists from 8th century, or the Homunculus of 16th century European alchemists such as Paracelsus, or perhaps even the Tulpa of Tibetan Buddhism's, and later Theosophists. ... To give you a current day timeline, a scientific paper that is often referred to as a breakthrough milestone is the Transformer's by Researchers at Google, published (only) in 2017. But let's take a quick detour... * https://arxiv.org/abs/1706.03762 * https://proceedings.neurips.cc/paper_files/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf * https://research.google/blog/transformer-a-novel-neural-network-architecture-for-language-understanding/ * https://www.wired.com/story/eight-google-employees-invented-modern-ai-transformers-paper/ ![What is _programming โ“_](./slides-export/6.png) ![What is _programming?_](./slides-export/7.png) Programming, AKA coding, gives computers precise instructions, very precise, which we call,for what you want them to do. ![How did you learn your mother tongue?](./slides-export/8.png) ![How _did_ you learn your mother tongue?](./slides-export/9.png) Quick show of hands... who a) ... who b) โ“ ๐Ÿ˜†