A quick glossary of our Generative AI Index use case definitions
Accent Modulation & Voice Augmentation
Some products focus on normalizing more distinct accents. Others have a broader focus on enabling a voice to be changed in a broad range of ways (e.g. make a male voice sound female).
Ad Generation & Brand Building
Products helping early stage brands further develop their brand identity. This includes AI generation of ads, of derivative marketing collateral, and of PR-style story pitches for publicity.
AI Editing for Video, Photos, and Audio
Companies are introducing AI to the editing process in myriad ways. For example, some products enable video editing from text commands or from a transcript, and others introduce diffusion models to a Photoshop-like image editor. The target end users range from zero video experience to professionals. Think of this as how generative AI expands the capabilities of the Adobe Creative Cloud suite.
Build Apps from Text Interface
This is similar to ‘Natural Language Frontends,’ but in a narrower problem space. Natural language LLM embeddings are mapped to specific tasks that can be completed using code-generating LLMs.
These tools empower business users to query databases otherwise accessed through SQL or Python with natural language.
Generation for Code, Terminal, and Querying
A number of companies are building on top of coding-, terminal-, and querying-specific LLMs. Particular focus is going into the engineer’s experience engaging with the underlying model. Unlike many business users, engineers are very aware that they can download an open source LLM trained on code and work with it directly, but are choosing to engage with an application that offers a slicker experience and features specific to their use case.
These are products built specifically for contact center operations. These largely help employees access knowledge faster or type a text response faster. Tenyx, meanwhile, seeks to fully automate call center operations. Call centers are also target customers for the Accent Modulation & Voice Augmentation category.
Draft and Pepper both started as SaaS-enabled marketplaces for accessing copywriting talent. With the rise of AI copywriting tools serving their talent pool, these companies have begun building similar AI in their own product. We expect incumbents will look to embed AI into their own products, especially where off-the-shelf foundation models are most applicable.
A key marketing best practice is to repurpose one ‘parent’ piece of media (e.g. a recording of a 3-hour panel) into content for various use cases (e.g. blogposts, Tweets, Instagram posts, TikToks, etc). These products leverage the strength of foundation models in matching genres to pick the key bits of content from a parent post and ‘re-jigger’ them into various other genres.
Tools offering design primitives and design assets.
There is perhaps opportunity to build an app around Markdown best practices with a wedge feature in AI generated documentation.
Figma to Code
These products generate code to replicate what is created in front-end design tools, like Figma.
Foundation Model Providers
These are companies building their own foundation models, often at the leading edge of research. Some additionally offer end user products.
General Purpose Communication Writing
These products feel like the next-gen Grammarly or GMail autocomplete. They are primarily browser plugins, not applications.
Generation for Copywriters & SMB Marketers
This is an early category of startups built on top of a foundation model. These products empower copywriters to engage with GPT-3 through a “skin” on top of OpenAI’s API that abstracts away some of the complexity of engaging directly with the playground environment. Some productized prompt-tuning occurs under the hood as part of the application layer. A number are now pushing into image generation and are also providing more workflow type tools for editing (a la Canva) and pushing content to the platform it will be published on.
MLOps, Infra, & DevTools
These products make foundation models accessible to software developers, not just AI engineers. They make it easy to select the right model, finetune that model, and deploy it.
AI generated music, which is typically rights-free.
Natural Language Frontends
This is among the most ambitious use cases for LLMs. The embeddings of a purpose-built LLM will map to tokenized actions that can be performed by the model (similar to how OpenAI Five was trained to play the video game Dota 2). Proficiency across a broad range of tasks will be enabled by a Mixture of Experts approach.
These companies apply diffusion image models to problems of visually branding and/or displaying products (e.g. putting clothes on fashion models).
These products play in the same broader category as Gong. They provide increasingly real-time and automed support & coaching.
These tools serve enterprise sales teams, helping their sellers do higher quality, more personalized, and more efficient work leveraging with foundation model enabled products.
These “search, reimagined” products take the theme of Google’s Knowledge Hub a step further. They use generative AI techniques to provide “the answer” rather than just links to possible answer sources.
Synthetic Talking Heads
It is newly possible to generate synthetic talking heads (although the work to offer such a product remains much more build-it-yourself than many of the other categories mentioned here). Use cases are still being explored by this cohort, but learning & development and personalized sales & marketing videos are two areas of early focus.
Synthetic Voices & Dubbing
It is newly possible to generate Synthetic Voices & Dubbing that sound almost as good as a live human voice. (The work to offer such a product remains much more build-it-yourself than many of the other categories mentioned here). These companies are productizing such models, largely thru an API offering, and sometimes with an app-layer product, for a variety of end markets.
Task Automation & Virtual Assistant
These companies offer RPA either directed through or automated through LLMs.
Tools for Artists
A catchall bucket for applications to help further (less-profit-oriented) creative processes.
These products generate synthetic data for training AI models.
While AI translation has been around for some time, LLMs enable it to happen on a more realtime basis, expanding the potential use cases.
This is a catchall buckets for products being built for specific verticals. Most tools remain horizontal.
Voice of the Customer Analysis
These companies use the embeddings of LLMs to bring quantitative techniques (statistical analysis) to an otherwise qualitative problem (understanding text-based feedback).
Today, many 1-800 numbers use voice-activated “call flows” as a first line of defense for directing traffic and weeding out some volume. These flows can be quite frustrating, given their reliance on keywords and the inflexibility of a rules-based approach. “Voicebot” products are offering a brand-new experience that replaces such flows with a combination of transcription, knowledge base access, text generation, and synthetic voice to power flexible conversation with a knowledgeable, problem-solving AI.
By applying LLM embeddings to a vector database, and matching these to the embeddings of plain language queries, semantic search is enabled as an alternative to keyword search.
Knowledge Bases / Semantic Search Apps
Semantic search infrastructure enables a new class of application for use cases like cross-app search and knowledge management.