On my flight home, I watched a university professor spend over 30 minutes grading a basic undergraduate business paper — still not finished by the time we landed.
An LLM could have handled 90% of the initial work in seconds, freeing the professor to focus on personalized feedback while the model learns and improves with each iteration.
The recent frontend updates to @github are making it unusable — on-page search is broken (always jumping to incorrect places), layout shifts everywhere, slow to scroll.
Ever wanted to mock the Cloudflare Image API locally? Here is a sample WordPress plugin that enables cdn-cgi/image/… routes locally and supports only image resizing.
Code signing doesn't solve everything because there are so many places where malicious code can creep into downloadable software. Source attestation is like a bill of materials with signed receipts.
Using browser cache for cookie-less user tracking is really smart. Assigning an identifier as Last-Modified or Etag response header and then tracking those subsequent requests seems unstoppable.
Is this why cross-domain requests have their isolated caches even for things like Google fonts, JS libraries, etc?
– An "agent mode" that allows chat and edit to work across all project files.
– The improved autocomplete now suggests the next edits you might want to make outside of the current line of code (defining additional class attributes, for example).
– With prompt files you can have an AI-specific documentation within your project to provide context for code preferences, architecture, etc.
Today I learned that Twitter allows changing password without asking for the second factor. Anyone with access to a valid session can simply change the password and lock you out. Same with account email apparently.
They do have this additional toggle for password resets (if email access is compromised, for example) but that is also disabled by default.
Created "Perf Bench" to generate all kinds of web performance tests. It is a single PHP file (inside Docker) that maps query args to feature flags. A helper script can generate all permutations of the supported feature flags which can then be used by any other benchmarking tool.