The Guide To Understanding Generative AI
2 September 2024What is Generative AI?
Generative AI (or GenAI) is artificial intelligence that can create content (text, imagery, audio) in seconds, on par with an experienced human and for next to nothing. It therefore has the holy trinity of time, quality and cost benefits.
In this guide, we will explore exactly how GenAI is affecting business, identify the key technological building blocks that make it happen and bring the technology to life by profiling a range of AI businesses, which Exactimo’s bootcamp alumni have either founded or currently work at.
How is Generative AI impacting business?
A lot has been made about how Generative AI will impact business. In fact, a 2024 study referenced in the Financial Times, showed that more than 50% of the US’s largest companies view artificial intelligence as a potential risk to their businesses, based on a survey of corporate filings that highlight how it could lead to wide-spread industrial transformation.
There are several other recent studies, such as this McKinsey report or quotes from company executives, such as Jared Spataro, VP of Modern Work & Business Applications at Microsoft who said “People take to the technology quickly when they see others around them using it, and once they have tried it out in their own jobs they don’t want to give it up”, which conclude the same.
But many of these studies talk in broad terms and in abstract ways, without getting to the nub of exactly how Generative AI is transforming companies. In this section, we’ll explore some real-life applications.
GenAI Use-Case 1: Content Generation
The most obvious workplace efficiency gain is content generation and the most impressive is on-demand video creation.
For instance, OpenAI’s Sora model can generate videos based on simple prompts, such as “create a video of a couple walking through a street in Tokyo with cherry blossoms to the side of the road” (see example ).
This level of content generation is already impacting creative industries such as film-making and advertising by reducing the time and cost of video creation.
GenAI Use-Case 2: Content Analysis
Another example is content analysis, where LLMs can review entire videos and pinpoint the exact timestamp in which a particular event occurred.
For instance, one example from Google Gemini, shows how artificial intelligence algorithms can identify in seconds the exact point in a movie in which a piece of paper is removed from a person’s pocket (see example ).
This will impact industries such as policing and law enforcement , as it means LLMs can process vast amounts of video footage based on useful human prompts, e.g. “show me when this man in video 1 first appears in videos 2 to 42”.
GenAI Use-Case 3: Product Discovery For Customers
Generative AI can also be injected into digital experiences in order to help customers discover what they need quickly.
Shopify, for instance, was one of the first companies to use OpenAI’s APIs to help its e-commerce clients improve their user experience. For instance, customers can converse with a chatbot, as if it’s a human, to find the right bbq for their needs. It then shortly released, Sidekick as well as other AI-based tools under the Shopify Magic umbrella .
GenAI Use-Case 4: Product Discovery For Companies
Generative AI can also be leveraged in order to help companies with their research and development efforts.
A recent example highlights how LLM helped Microsoft and the Pacific Northwest National Laboratory, which is part of the US Department of Energy, found a new chemical substance which is like Lithium , potentially helping to reduce the earth’s dependence on lithium mining following the surge in electric vehicles.
GenAI Use-Case 5: Software Augmentation
Software tools have also become more sophisticated with the integration of LLMs. One of the first examples was Slack’s use of OpenAI, which enables employees who miss a Slack call to get a summary of what was discussed and an automatically generated set of meeting notes. It can now do so much more , such as search projects, teams and topics to find what you need with questions such as “what’s the Q1 sales strategy”.
Another example was profiled at Apple’s 2024 product launch that demonstrated how LLMs, which are now built into iPhones, are multi-modal in being able to piece together information from multiple sources to help users journey plan, such as this user’s need to help her pick someone up from the airport.
GenAI Use-Case 6: Market Research
Generative AI is being used to improve workplace productivity through its lightning-fast market research capability.
Product Managers, for instance, are using AI-powered search engines like Perplexity to help research successful product strategies from other companies, which they can learn from.
Or wealth managers at Morgan Stanley are able to leverage digital assistants (powered by OpenAI), which are trained on its vast knowledge base to better serve customers .
GenAI Use-Case 7: Customer Support
Generative AI is also being used to make customer support more efficient by powering bots that can handle customer support enquiries.
Klana, for instance, has recently turned a profit for the first time and the CEO credits AI for fuelling its path to profitability through intelligent customer support assistants . Klarna claims that the technology has boosted the average Revenue per Employee by 73% over the last 12 months, rising from SEK 4.0m (€350k) to SEK 7m (€610k).
How does Generative AI work?
It’s worth noting that Artificial Intelligence has been around for decades. You can see this in the search terms on Google Trends, which demonstrate comparable search volume back in 2004.
Its decades-long existence is also apparent in books such as If Then , which trace the use of clever statistical simulations powered by huge computers, run by organisations in the 1960s that worked with the US government to predict and prevent race rioting and the spread of communism.
But in order to explain how Generative AI works and its recent eminence, it’s worth breaking down the key components that have made this subset of AI emerge as a significant technology.
To make it easy to understand, we like to describe Generative AI as having three core technological components, one of which is the engine and the other two the fuel.
The Engine: Large Language Models
The engine that powers Generative AI is the Large Language Model (or LLM), which is trained on massive amounts of data. This data can include books, articles, code, scripts, and more. By analyzing this data, the LLM learns patterns, grammar rules, and factual information.
The other characteristic is that LLMs can understand and communicate like a human, so can give helpful answers to human prompts, making the user experience feel native.
The Fuel I: Computational Power
The fuel that powers the engine (LLMs) is computational power, as measured by FLOPs, which is a metric used by computer scientists to measure the processing capabilities of computers, which are used to train AI systems.
The compute used to train AI models has increased by a factor of one hundred million in the past 10 years, which means we have gone from training on relatively small datasets to training on the entire internet.
This is why there have been several legal cases against AI companies like OpenAI from the likes of New York Times, Reddit and Scarlett Johansen who claim that OpenAI, through the use of supercomputers, have trained its models on their content. And as computers get stronger and faster, the legal wranglings will only increase.
The Fuel II: Vector Database
The second key fuel that powers the Large Language Model is the Vector Database.
Vector Databases allow software to easily compare and find similarities between things via Vector Embeddings, which are lists of values that quantify certain characteristics or features of items, e.g. this word is like that word, or this song is like that song. In practical terms, a vector database, automates the structuring of unstructured data.
Prior to a Vector Database, humans would have to manually tag or define items in a database to tell the computer that a particular item was or how it looked. For instance, an apple would be marked as being ‘red’ in colour, ‘round’ in shape; a banana is ‘yellow’ and ‘long’, whilst an orange is ‘orange’ and ‘round’.
With this new type of database that programmatically applies vector embeddings to things, a LLM can rewrite a paragraph in seconds as it knows that the word ‘perpendicular’ is the same as the phrase ‘at right angles’ or software programs like Spotify can tell you that Colter Wall’s songs sound like Bob Dylan, so if you like Bob Dylan, you should like Colter Wall or that a chemical substance is the same as Lithium (see example in previous section).
How can Generative AI Augment human creativity?
Generative AI is able to augment human creativity by leveraging Large Language Models to create content, analyse content and boost workplace productivity based on the requests and preferences of individual users.
Which technology helped Generative AI create convincingly authentic media?
It’s the combination of Large Language Models, powered by supercomputers and vector databases that has enabled AI to create convincingly authentic media. A consequence of this capability is the exponential rise of Deepfakes , due to the increasing adoption and greater sophistication of AI models.
What is the difference between Generative AI and LLMs?
Generative AI and LLMs (Large Language Models) are often used interchangeably, but there are key differences.
GenAI, is a broader term encompassing any AI system capable of creating content (text, imagery, audio) in seconds, on par with an experienced human and for next to nothing. Whereas LLMs is a key engine that is trained on massive amounts of data and can perform tasks like translation, writing different kinds of content and answering questions in an informed way.
More examples of Generative AI in business
In this section, we’re going to bring the application of Generative AI to life by profiling a set of AI-based companies, which attendees from our bootcamps have either founded or worked at.
What does Stability do: Stability AI develops models for generating images, text, and other content from textual descriptions. Its most notable contribution is the Stable Diffusion model, which can generate high-quality images from text descriptions.
Exactimo Alum Name and Role: Joséphine Parquet - Product Manager - MiM at London Business School
—
What does Captur do: Captur is changing how fleet operations & management is conducted for Delivery, Micro-mobility and Automotive companies. Captur replaces manual reporting with smart visual inspections, using industry leading AI to automatically detect issues and create tasks.
Exactimo Alum Name and Role: Charlotte Bax - Founder - MBA at London Business School
—
What does MyAsk AI do: MyAsk AI builds AI customer support agents for SaaS companies, trained on websites and internal documentation that are cheaper, faster and better.
Exactimo Alum Name and Role: Alex Rainey - Founder - Undergraduate at Bristol University
—
What does Qantev do: Qantev is revolutionising the claims process and fraud detection within health and life insurance by using artificial intelligence that automates the decision-making process of a medical expert through all stages of a claims journey.
Exactimo Alum Name and Role: Claudio Randolph - Strategic Project Manager - MBA at HEC Paris
—
What does Sphere do: Sphere uses LLMs to help companies put global indirect tax on autopilot (sales tax, VAT and GST compliance) - handling everything from monitoring, registration, calculation and filing in one click.
Exactimo Alum Name and Role: Nicholas Rudder - Founder - MBA at London Business School
—
What does Encord do: Encord is the leading data development platform for advanced vision and multimodal AI teams, building tools and infrastructure to help the world’s leading AI teams get their models into production faster - with data-centric model testing, human-centric workflow and annotation tools for labeling.
Exactimo Alum Name and Role: Matt Tang - Client Ops Lead - Oxford MBA