Discover more from US|AI Blog
An AI platform for sourced knowledge
This week we launched Abbey, an AI platform for sourced knowledge. We want Abbey to help you learn anything: pick up new skills, explore liberal arts, or just develop your curiosity. It should not take a degree and years on the job to reach a minimum level of competence in one subject: we hope that Abbey can make everyone an agile learner -- to get up to speed at a job, to get ready for a new career, or just to have fun.
We are building Abbey with these two principles at heart:
High-quality sources, like academic articles and textbooks, are key to effective learning.
Abbey is not pedagogically prescriptive: we want people to use Abbey as they might want to use a library.
We don’t want to just regurgitate the first page of google search results; search engines are too far gone. We also want your learning to be structured but ultimately self-guided.
Who’s Abbey good for?
Students who want to do their reading more effectively, or who want to better understand their course material.
Researchers who want to dive deeper into papers.
Professionals and companies who want to put their own documents into Abbey to help everyone get up to speed or build their skill-set.
What Abbey is today and what we want it to be are two different things. Below we’ll discuss where we are and where we’re going.
It starts with a good way to read your documents but ends with an entirely new way to learn.
The Best Way to Read
Reading is hard: we miss important content, misunderstand things, and often can’t find what we know to be in the document. Abbey makes it much easier to read actively. Here’s how I use Abbey to read:
Summarize: every document uploaded to Abbey comes with a summarization tool. The generated summary can take a minute or two to build since it is carefully examining the entirety of your document, no matter how long. When I first try to read a document, I use Abbey to make a summary, and then ask it to shorten the summary.
Q&A: After I read the full and shortened summaries, I start asking questions about the parts I find interesting.
Showing sources: Each response to a question yields citations to specific passages in the document that I can examine to make sure the response is accurate. I also like to get the full context; this is how I begin to read the actual document.
Semantic search: if I want to find more places in the article to read, I search over the document using our semantic search tool. This search is more like Googling over your text than using command-F (or control-F); it’s more powerful and can understand concepts and synonyms.
These features are available with every document you upload: Word docs, Excel files, and even PDF scans that are not ordinarily searchable. Abbey uses optical character recognition, a technique for recognizing text in images. The same goes for screenshots you upload.
As the complexity of our tools grows, we become less knowledgeable about how they work. We hope that the features and interfaces we provide on Abbey help you better understand how the platform works and ultimately help you use it better. We don’t believe in keeping functionality from you under the guise that it’s proprietary knowledge.
Use your documents: Users can upload anything they want to Abbey, and it will be there when they get back. You can organize those documents into folders so that you can use Abbey like a file system.
Choose your model: we give users a choice between different models, including OpenAI’s GPT-4 (for subscribers). We hope to add even more models in the future.
Choose your randomness: If you have used other chat-bots, you might notice that asking the same question a second time can yield a different response. We give you control over how much randomness is involved in this process: do you always want what AI thinks is the best answer, or do you want to try out different ideas?
More: In the future, we want to give users even more control over how Abbey works. For example, we plan to give users control over the specific prompts used to generate summaries and responses.
Some things aren’t available yet on Abbey, but they will be soon. We’re working hard to make Abbey a much deeper offering than other platforms.
Content Library: Textbooks and papers will be available for free and for purchase on Abbey to add to your library.
Generated Curriculum: We want learning to be structured and personalized. Using sources you upload as well as our library of content, we want Abbey to outline a set of AI-powered assets to help you achieve your learning goals.
Much of the current AI discourse centers around its capacity to make our lives easier, but this is shortsighted. Masking and simplifying complex jobs in the hope that AI will handle it for us forces us to imagine a future where nobody has to do anything.
In reality, AI removes the ceiling on human potential; it allows us to do harder things that we couldn’t previously do alone. We’ll need lawyers, engineers, plant operators, critics, teachers, parents, and more to all have a better understanding of our world and the tools in it. We could flounder in pessimism, sure that the jobs will all be gone in ten years. Or, we could bet on ourselves and use AI to raise the floor of human competency.
This is not to say that Abbey is difficult to work with; it’s straightforward. The point is that as we iterate, we’ll work to provide the tools you need to learn the thing you thought you never could. Deep and accurate content libraries, curriculum generation, knowledge assessments – Abbey is a bicycle for the mind. These things are all on their way, and you can use them however you’d like.