By the time you are reading this, I will be on a 🚢 to Alaska. So I’ll be taking a one-week break to focus on recharging. I’ll be back in your inbox next week with a fresh batch of content.
Software Engineering ⚙️
ts-migrate
My coworker and I made our Typescript stricter on our monorepo with more than ten thousand Typescript files. We had more than twenty thousand type errors, and it would have been too painful to fix in one go if we did not have this ts-migrate script from Airbnb. They developed this script to automatically ignore type errors so that they can gradually fix the type errors.
Meet New Relic Grok, the first GenAI assistant for observability
New Relic says I will soon be able to ask, “What percentage of our users failed to complete the campaign creation workflow?” instead of having me go back and forth between our code and New Relic to conjure up a SQL query. I will propose that my team try this tool as part of our dashboard review sessions. My first question will be, “What should we look at first?”
Learn how to build a custom test reporter using node. js’ new native test runner
I did not know that Node.js after v18 comes with a native test runner. It was also interesting to learn that you can either stream or batch the test report generation.
Koala Evaluation Set
This evaluation set was used to test the effectiveness between two large language models by feeding the questions to the models and having humans evaluate the responses. The set is based on the actual user prompts, so I enjoyed reading some of these prompts. One of my favorites was “Write a poem about Mike and Joe becoming millionaires by leveraging the power of AI to become the greatest Agile coaches in history. Include content from the agile manifesto.”
Business đź’°
Google “We Have No Moat, And Neither Does OpenAI”
A leaked internal document of Google’s claim expresses concerns that Google’s competitive advantage in AI, the almost infinite amount of money to train the model, is not enough. Just as stable diffusion overtook Dall-E, an open-sourced large language model can overtake ChatGPT.
The Unified Content Business Model
Just as you manage a sale funnel for a product from awareness to consideration to conversion, content businesses also have multi-tiered touch points with consumers, from advertisements to subscriptions to offline consumptions.
Context SDK – Introducing the most intelligent way to know how and when to monetize your user
This SDK lets an app classify its user’s context (running, sitting on a couch, etc.) to provide the right experience for that context. Theoretically, you can do the same on mobile web as well as you can run ML models with WebAssembly, and can get access to the same sensors. But getting user permissions for those sensors will be much more difficult.
youngk.im 