Intelligent Document Understanding Explained

Get the guide to data for GAI → ibm.biz/BdmS2v
Learn about related technology → ibm.biz/BdmS2K
Intelligent document understanding is when technology is used to help a subject matter expert read, understand, and make a decision based on data. Brandon Swink dives into how this works and the architecture behind it.
AI news moves fast. Sign up for a monthly newsletter for AI updates from IBM. → ibm.biz/BdmS2a

[@tomoki-v6o]

Llm generates a ton of information .for one prompt . And only a small percentage of its usefull for the user. if the user can replay and highlight (points directly interactively ) to the important parts .that improves the user experience . and helps the model to improve. In the current interaction the models forgets or cant understand what the user means. you have to prompt it every time

[@mzimmerman1988]

nice work 👍

[@physical8.5]

nice Ibm best technology and science group!

[@duoko98]

Also, to highlight, As domain matter subject documents become more dense in terms of mathematical, numerical, statistical, graphical, or scientific content or in terms of the number of subchains of dynamical, phenomenological, and real world context specific strings of thoughts or language sentences required to understand each sentence, paragraph, etc and the overall meaning and conclusion (explicit stated or implicitly derivable conclusions because both are needed for end decisions in scientific contexts ); The machine learning algorithm as a large language model needs to be able to properly answer very advanced and domain specific questions about the content it is trained on in order to be efficient we need to move past it only being able to summarize and mostly just regurgitate the paper… to being able to answer much deeper questions:::: We should be able to do things like derive all explicit and implicit physical-logical-dynamical assumptions required to make a scientific document entirely true and ask the AI fundamental questions related to the network of things that have to be true or are true to even produce the statements in the paper itself we can learn more this way… Or for example let’s say we want to do an in depth cross analysis on the treatment of the teaching of the subject of electromagnetism across 3 different authors in order to learn more about the how the differences in ways the subject is taught highlight and reflect the nature of our own perception of the fundamental phenomena and how that can help us learn more -> the AI should have that capability::::: or let’s say we are reading a highly quantitative document -> we should be able to use the AI not only to query what those quantities are and what they are described as in the paper but we should also be able to cross reference this data with other data in scientific repositories effectively allowing us to run advanced numerical-logical queries without ever having to be in a lab because not only does the AI fully understand the scientific domain it also fully understands how to apply analytical/and functional mathematical methods to different repositories of data… --> I think I could list like 10 more related applications for scientific AI that would help people do things like reach faster conclusions on the nature of quantum mechanics or maybe there’s a way to use Machine Learning to do an knowledge based optimized brute force search for a solution to the Riemann Hypothesis through a deep analysis through the maths etc etc

[@stanTrX]

Whats the difference with RAG?

[@justinchen207]

I’m watching this on like triple speed. Can’t they just make the graphics and slide before hand and actually get more than a paragraph worth of info in a 10m+ vid

[@_hadoken]

Yeah I like these videos and what they aim to do but I'm not a fan of this format. Even the speed aside, there's always bad handwriting or silly errors like this one using a possessive apostrophe in SME's when it should be SMEs, and "INCREAS". But it's hard to think on the fly with little chance to practice, I get that. It would be easier for them to prepare the visuals instead.