What are Vector Embeddings Let’s go back to the number line. The distance between two points; This is a good example of what Vector Embeddings are, fingerprinting a document into a number in multi-dimensional space. Since a document can be represented as a number (series of numbers), now a relation can be made between two documents. The relation in terms of distance between two vectors (which could be a markdown document for instance), small distances have a high relatedness and large distances suggest a low relatedness....

Saw an ask on Twitter A person I follow on Twitter tweeted: Is there a git/version control for prompts? I hate losing great prompts and I’d love to be able to iterate on them. So I created a quick command line chat to GPT-4 https://github.com/dathan/go-openai-prompt-git-save It is a go program, and you need an OPENAI Token and a GitHub User Token in your environment. The program is very simple. From the terminal, have a chat interface to OpenAi, and save the prompts asked, to GitHub, with the content from GPT-4’s response....

What is Prompt Engineering Let’s ask chat gpt: Response in text Prompt engineering is the practice of crafting effective prompts to generate specific outputs from language models like LLMs (Large Language Models). It involves selecting the appropriate vocabulary, formatting, and instructions to help the model understand the desired task and generate the desired output. Effective prompts can improve the performance of LLMs in various natural language processing tasks, such as language translation, sentiment analysis, text summarization, and language generation....

What problem am I trying to solve Extract content from a dailyco + deepgram session realtime What am I trying to do to solve the probkem Display the transcription realtime display the user and what they said display the avatar The avatar has a tooltip of the username make the body transparent and lighten up the boxes Summarize the transcription like a CNN key takeway of a speech. buffer the messages, once the messages hit a threshold empty the buffer AI Summarize fill the display buffer What is the priority Display –> 1d and will require big changes Summarize –> 1d an minor changes What am I going to do scope this to development ✅ display the avatar ✅ on meeting join and isOwner start transcribing ✅ show the transcription button for clients ✅ Just finished the design buildingof a real-time transcription summary service for meetings....

Motivation As an Engineer, I’m always on the lookout for new and exciting challenges. So when I was tasked with solving a rendering problem in a Typescript application, I jumped at the opportunity to learn and figure it out. In this blog post, I’ll be sharing my experience and insights into rendering in a Typescript application within a Vercel environment, using Node 18+, Next.js, strict Eslint guidelines, and Chakra UI for custom components....

Motivation I have been writing code to support sharing content from the web on various platforms since 2010. This is my attempt at recording all the things that I do to effectively share effective Previews across LinkedIn, Facebook, iMessage, Twitter, and any other Open Graph compatible system. History The Open Graph protocol was first introduced in 2010 by Facebook. It was created as a way for website owners to better control how their pages were represented when shared on Facebook....

Goal Record cloud video and transcribe it Tasks Setup cloud storage endpoint ✅ Build a meeting token endpoint and pass that to the call creation ✅ enable recording - when the session owner joins ✅ add command line applications to the base – not going to do will use cloud functions download file wrapper - not needed will create a signed url process file wrapper upload successful process wrapper update database state run command line from github doc the results and how to reproduce

Integrate vendor package and give results remove API from the front end ✅ save? room and return the room id that is created ✅ creator makes rooms ✅ participants need to be a part of a session to join a room ✅ need an error layer ✅ verify session against joining a room ✅ fix the buttons to look like a poor man’s zoom in terms of layout ✅ Add speaker view to the layout add new button layout using their button scheme add code to add fake cells

As software engineers, especially product engineering, one of our core objectives is to deliver software solutions to customers quickly, while still maintaining the quality of the product. It is not enough to simply produce a product and ship it out - we need to ensure that the product fits the customer’s needs, and can evolve to meet the customer’s changing demands. To achieve this, we need to establish a set of software delivery practices that prioritize speed and iteration to find product market fit....

OpenAI fine-tuning with a notebook in VSCode AI is the future. Co-pilot is the future. Building a finetuned layer for GPT-X will enable products to produce results faster, with less upfront effort for a fraction of the cost. Setting up a training set Let’s assume for this post, is based on some structured data. That structured data has been tagged by a human, and there is a lot of it. Roughly 3 MB of data....