–
- A few lessons from Crowd strike incidents for engineers: Quantify the impact of crashes before releasing, do canary (staged) deployments to minimise risks, treat all changes—code, configurations, and infrastructure—with equal importance and manage them through a unified system, monitor dependencies and vendor updates to avoid unexpected disruptions. The biggest-ever global outage: lessons for software engineers
-
Open-source AI allows developers to customise models with their data. It provides independence from closed vendors, improves data security by enabling local use, and is cost-efficient. Its rapid advancement makes it a strong foundation for long-term development. [Open Source AI Is the Path Forward Meta](https://about.fb.com/news/2024/07/open-source-ai-is-the-path-forward/) - Airbnb uses reusable infrastructure to progressively roll out different versions of React across their repository, measuring the results of upgrades. They also uses module aliasing and environment targeting (production vs. treatment production) to upgrade incrementally and test effectively.
- Our enemies are not bad people – but rather bad ideas. The techno-optimist manifesto.
- Entropy is the amount of information we don’t know about a situation but could learn. What is Entropy?
- Europe does a bunch of cool things (and they are not apps): Chocolates, Whisky, Agriculture (Dutch), Chips (ARM), SWIFT (the bank thing), drugs to cure diabetes, R&D for the aerospace industry, Formula 1, Supercars, Superyachts, Gene sequencing, big science projects like CERN and extremely large telescopes.
- Ideas and tips for reading books
- The main function of a book is to act as training data for the LLM inside of you.
- When reading a book for the first time, you’re doing two distinct jobs: learning from it and making it more user-friendly for the future.
- You don’t need to finish every book you start. Develop the skill of not finishing okay books and bad books. Spend most of your time with great books.
- The three trending SaaS business models in AI land
- Seat-based – add extra dollars per seat (Slack, Notion)
- Usage-based – charged based on usage (Mailchimp, OpenAI APIs)
- Outcome-based – charged based on the value provided (seems hard to track?); this is more like agents doing the work that you would have normally shared with someone on platforms like Fiverr (https://agent.ai/)
- Some engineering news
- React Native Filament allows you to load GLB (3D) objects on mobile phones
- Node.js officially starts supporting Typescript
- Minisearch is a smart frontend search
- It’s easy to use HTTPS with localhost
- Some AI news
- Mixtral launched a Large Enough LLM
- Facebook launched 405B parameters LLM — apparently better than Claude/GPT-4o (?)
- ChatGPT launched their most cost-efficient intelligence (mini) model GPT-4o-mini
- OpenAI launched SearchGPT – their foray into search business – might kill perplexity first