A Beginner's Guide to Gen AI
Colin Levy
Director of Legal @ Malbek - CLM for Enterprise | Legal Tech Author and Speaker | Fastcase 50 2022 Honoree | Legal Tech Startup Advisor and Investor | Mental Health Advocate
We've all heard the term Generative AI. Many of us have experimented with it or used it. BUT, what exactly is Generative AI? In this brief guide, I will introduce you to what it is and why it matters.
Generative AI uses algorithms, specifically machine learning algorithms, to review large amounts of data and then receive outputs based on this large input of data. The "Generative" part of Generative AI refers to these tools' ability to create new content based on the data it has been trained on.
Before getting into use cases for Generative AI, let me distinguish Generative AI from "traditional" AI. Traditional AI uses predefined rules and code to perform specific tasks pre-defined for it. Generative AI has no predefined tasks. You can define what Generative AI does based on the data it has been trained on.
There are different types of Generative AI models. I want to briefly touch upon each type:
1) Generative Adversarial Networks: Known as GANs for short, these are used to create realistic images and videos.
To illustrate, consider this. GANs behave like two people competing against one another. One looks at something and decides if it is real or fake. The other provides feedback. The game continues until the evaluator is able to more accurate distinguish between the two.
2) Transformers: I'm not talking about the movies or the toys, though I did love both. These are models used for creating predictive text and recognizing speech.
For example, you are writing a text and your phone starts to finish your message. This is an example of a transformer.
3) Variational Autoencoders: All you need to know is that these models are used to create artificial images and text.
For example, you give your friend a picture. The friend takes the picture and makes it smaller to fit into a certain frame. But then you decide you want the original photo back. Your friend...who has been watching Star Wars and somehow understands The Force, is able to turn the photo back into its original size. This is what this type of model does.
4) Flow-Based Models: These are also used to create text and images, but in a specific way.
领英推荐
For example, consider you have a certain number of legos. No more and no less. You can create whatever you want with that number of legos, but cannot add more or use less. This is how a flow-based model works to create things.
5) Recurrent Neural Networks: These are used for speech recognition and natural language processing.
Say two friends are playing catch. One of your friends is an excellent athlete and is able to always catch the ball seemingly without moving an inch no matter how it is thrown. This type of model behaves like this friend. It can do something consistently based on all previous activities of that type.
Okay, so now that you know the types of models thet exist, it's time to touch upon the concerns and limitations of these models:
While AI itself is not a new concept or technology, Generative AI is. The growing impact of Generative AI cannot be overstated or underestimated. How the world works is changing. The way forward requires three things from each of us - a) A willingness to learn and to adapt, b) Being comfortable being uncomfortable at times as Generative AI evolves, and c) Recognizing that technology is here to augment our own abilities, but to do so effectively requires experimentation and iterating.
This article is just a high-level and brief introduction to this exciting space. To learn more, I strongly encourage you to check out these resources:
1) Josh Kubicki's excellent newsletter, The Brainyacts.
2) This article from ZenDesk (be mindful of its customer-service focused bent due to Zendesk being in that space).
3) This excellent resource from the Michigan Institute for Data Science at the University of Michigan.
4) This interview with the Dean of Suffolk University Law School, known for its legal tech prowess.
Hey Colin! ?? It's fascinating to dive into the world of Generative AI, but remember, as Albert Einstein once said, "The true sign of intelligence is not knowledge but imagination." So, exploring AI could also unlock new ways of thinking and solving problems in legal tech! If you’re interested in combining innovation with environmental impact, Treegens invites you to check out an upcoming sponsorship opportunity for the Guinness World Record of Tree Planting ????: https://bit.ly/TreeGuinnessWorldRecord. Your insights could contribute significantly! #innovation #environment #legaltech
??Data Storytelling Strategist ?? Author ??Emerging Tech Influencer??ABA Women in Legal Tech 2023 Honoree????NH 2024 most influential business leaders??Tedx Speaker?? Inclusion advocate??Keynote Speaker
8 个月Fantastic breakdown and easy to understand Colin Levy, well done.
Professor of Computer Science and Law, Emeritus, at Rutgers University
8 个月I would recommend the resources listed in this post by Dan Hunter and those discussed in the comments. https://www.dhirubhai.net/posts/dan-hunter-ab8803131_generative-ai-exists-because-of-the-transformer-activity-7156683966081572865-GuTM?utm_source=share&utm_medium=member_android
CTO at TermScout
8 个月Nice overview Colin Levy. Another point I like to make about AI, especially in legal tech, is that machine learning and AI are not at all new in this field. Much of the recent excitement is around the generative piece. Reviewing contracts with AI, extracting defined terms, identifying clauses, redlining, named entity extraction, deviation analysis, etc. are decades old techniques and most CLMs were using under the hood before the current hype cycle.
Director of Legal @ Malbek - CLM for Enterprise | Legal Tech Author and Speaker | Fastcase 50 2022 Honoree | Legal Tech Startup Advisor and Investor | Mental Health Advocate
8 个月What do you think of my guide and/or what are other resources you would suggest checking out?