Daniel Bigham Ideas

Idea: Using data from Facebook and/or Twitter, extract a trust model/network that specifies how trusted people are on different things, akin to Google's PageRank algorithm. Then use that trust network to poll people on important questions. Related: Governments, decision making

Idea: Fine tune GPT-3 on the Bible and highly regarded Christian books, and then have it take each chapter/verse of the Bible and create a new version/translation that infuses artistic/creative spirit akin to The Message.

Idea: AI model where you feed it a bunch of information about a family member or friend, including fun memories, and then it produces a delightful song about the person in the requested style.

Idea: Use ChatGPT (etc.) for language learning. i.e. A specialized mode where it would act as a language tutor, talking to you in the target language in a gentle way, building up your skill.

Idea: Synergy between making liquid oxygen for rocket launches and taking CO2 out of the atmosphere: Cool air down until you get liquid oxygen, and along the way, capture the solid CO2 that forms.

e.g. Where would the liquid oxygen actually come from to fill the trucks that deliver it to the launch site? (You cool air down to the boiling point of oxygen, so it becomes liquid) e.g. What is the boiling points of the gases in air? e.g. Why is CO2's boiling point so high?

Idea: App for creating smart macros where you give it one or more recordings of doing a task, along with a textual description with each recording. You can then give it a new textual description, and it does the UI clicking/typing to accomplish the task. @Microsoft @AutoHotkey

Idea: App that reads whatever text is on your screen every 5 seconds, produces embeddings for it, and then allows you to search that information to easily get back to places you've been, regardless of what app it was in. @Microsoft

Idea: On Twitter, have a button you can click that means "I wish I didn't see this". Twitter then trains an AI model just for you to better filter out tweets and hide replies that you'd prefer not to see. Limited to blue check mark subscribers. @elonmusk

Idea: AnswersGPT: You ask it a question, and it has models that generate related questions that it can ask Wolfram Alpha, Alexa, Google Assistant, Google search, etc. It feeds the results back into a ChatGPT-like model that reads those Q&As, and then synthesizes a final answer.

Idea: AI model that takes as its input a Jira ticket and an existing code base, and produces a pull request to implement the requirements of the ticket.

Idea: Once AI is able to "watch" movies and produce rich embeddings for them, produce a movie recommendation engine on top of those embeddings. i.e. The user rates 100, movies, and the engine uses their embeddings to produce a good model of the user's tastes.

Idea: Create an AI model to review pull requests and make code review suggestions. Allow users to thumbs up/down suggestions, so it can improve over time. In addition to using this for human-written code, also use this to improve machine-written code.

Idea: Use the @OpenAI Embeddings API to produce an embedding for every sentence, paragraph, section, and document in all of your emails, documents, meetings, etc. Then, be able to highlight any text on your screen, right click, and get a menu to jump to the nearest other things.

Idea: App where you take a song, as well as a recording of your singing voice, and it replaces the singer's voice with your own. (even if the singer recorded multiple tracks of them singing harmonies)

Idea: An AI model where you give it a song and it decomposes it into its sub-tracks, produces a digital instrument for those tracks, as well as sheet music.

Idea: A drone that flies around your home while you're not around to create a digital model of where everything is, allowing you to later ask questions like "Where is Emma's teddy bear?".