Unification of relational data and vector storage sounds undeniably compelling for semantic search usecases, but it’s important to look at the tradeoffs before going all-in on such an approach. The article’s ‘The Limits of This Approach’ section explains it quite elegantly.
Parts of this article on tokenmaxxing read like plots out of outlandish worlds of Cyberpunk and Mad Max. I mean, didn’t we already learn our lessons from ‘lines of code as a performance metric’?
I can see a bit of myself in all 3 rejected candidate types. What helped me crack interviews was doing more of them (in ‘production’, of course) and learning from failed ones. It was 10x more effective than any mock interview I ever did.
Glad to know GH approached agent security from a position of distrust rather than trust. Remember what happened to Sonny and VIKI in iRobot despite the strict enforcement of Asimov’s Three Laws of Robotics. You remember it, right?
Aren’t companies firing employees to manage rising AI costs, among other things?
Nothing earth-shaking, TBH. Spotify’s release process is mature and grounded but resembles how releases are contemporarily done in most large product orgs. At Deel, we had a strikingly similar GTM process for new features.
On a good day, I nail my introduction and come out impressively. Sadly, good days are far and few. Steve’s advice to write down your introduction to be always prepared sounds pragmatic. This is going in my urgent TODO!
Searching by intent (aka semantic search) is the antithesis of searching by keywords. Amazon’s semantic search solution is pretty amazing at how it leverages mega LLMs to (a) build knowledge graph to serve known queries, and (b) create smaller LLMs for generative responses to serve never-before-seen queries.
AI token usage in performance reviews? Seriously?!
Career growth in 2 sentences - ‘Your company will never tell you that you’ve gotten too comfortable because they benefit from your comfort. You have to be the one who notices it and breaks out of it.’