On Saturday, November 08, 2025, at 3:55 PM, James Bowery wrote: > While It's ok to "vibe code" with these things with the understanding that > you're not getting much out of them beyond Wikipedia BS tailored to your > biases, in order to actually make progress you need run code that's been > written to test-as-spec. As I've previously pointed out, even then they'll > try to get away with shit by constructing "tests" that report they pass > without testing anything, but the thing is that's so easy to catch that it's > kind of cute of them to try.
They ration resources in various ways. Sometimes you can get it to do all kinds of work other times not. There are systems though where you can have the AI build code via spec alone for example using Jira work items. And more resources are coming fast with the rush on new data centers and tech like: https://www.livescience.com/technology/computing/china-solves-century-old-problem-with-new-analog-chip-that-is-1-000-times-faster-than-high-end-nvidia-gpus and UltraRam: https://www.tomshardware.com/pc-components/ram/ultraram-scaled-for-volume-production-memory-that-promises-dram-like-speeds-4-000x-the-durability-of-nand-and-data-retention-for-up-to-a-thousand-years-is-now-ready-for-manufacturing ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T19f9264b814f3dfa-Mc50678f8eab95a63fe9c5e4f Delivery options: https://agi.topicbox.com/groups/agi/subscription
