Musk Says Grok Bot Was Costing Too Much to Run!

10 stories in this episode

1. 01

   Many Grok Bot users' weekly meters hit 100% early; Musk: serving it cost too much

   Grok Bot subscribers found their weekly usage at 100% days before the reset; when one asked for a reset, Elon Musk replied, "The cost of serving Grok Bot was unsustainably high relative to revenue," without saying whether limits will change.

   0:00Read the original · X ↗
2. 02

   From December, ChatGPT and Codex draw from one token allowance

   OpenAI's help page says Plus, Pro and Business plans will share one token-based allowance across ChatGPT and Codex from December, with Instant chat exempt; it hasn't said how many tokens each plan gets.

   1:29Read the original · OpenAI Help Center ↗
3. 03

   OpenAI plans to call its data centers "AI compute campuses"

   Axios reports OpenAI will start a policy and PR push after the midterms, pushing AI "speed limits" (mandatory safeguards for the most powerful systems) and calling its data centers "AI compute campuses," a name an OpenAI spokesperson confirmed to Business Insider.

   2:33Read the original · Axios ↗
4. 04

   "Mythos 5.5" sat in Anthropic's page code; "gpt-6.2-astra" showed up in Codex usage stats

   Hidden markup on Anthropic's Mythos page read "Claude Mythos 5.5" on Saturday and said 5.1 again by Sunday afternoon, while some Codex users saw "gpt-6.2-astra" in their usage stats; neither company has commented, and a "leaked" Mythos-class model on Hugging Face turned out to be fake.

   3:41Read the original · X ↗
5. 05

   New benchmark GDPx: of 20 models tested, the best, Opus 5.5, scores 12%

   SemiAnalysis launched GDPx, 206 long agent tasks across 11 high-value fields; of 20 models tested, Claude Opus 5.5 leads at 12%, Gemini 4 Carbon scores 10.9% at roughly a third of the cost per task, and SemiAnalysis notes safety filters pull down Opus and GPT-6 Astra in three domains.

   5:10Read the original · SemiAnalysis ↗
6. 06

   Claude found a Linux bug and wrote a root exploit for it; Red Hat has patched, Ubuntu is still in progress

   Anthropic's Frontier Red Team says an unreleased internal model found serious bugs across major software, including a Linux kernel flaw (CVE-2026-98432) Claude turned into a remote root exploit; the kernel is fixed upstream and Red Hat has shipped patches, while Ubuntu's are still in progress.

   6:41Read the original · Anthropic ↗
7. 07

   Karpathy had Opus 5.5 build a YouTube from nearly 20 years of HN picks, in about 2 hours

   Andrej Karpathy had Claude Opus 5.5 sift Hacker News comments from 2007 on into 6,454 recommended YouTube videos and wrap them in HackerTube, a login-free YouTube with middle-frame thumbnails, in about two hours.

   8:20Read the original · karpathy.bearblog.dev ↗
8. 08

   itch.io's new game pages doubled; an indie launch with 11,000 wishlists peaked at 15 players

   itch.io says new game pages per day doubled in two months while downloads didn't, one tracker's chart says about half of October's Steam releases disclose AI, and an indie publisher blames an AI-clogged review queue for a late launch that peaked at 15 players at once.

   9:41Read the original · itch.io on Bluesky ↗
9. 09

   Kid Rock says the gospel choir on his Eminem collab is AI

   On Real Time with Bill Maher, Kid Rock said the gospel choir on "Heal My Faith," his song with Eminem, was made with AI prompts, defended it with the drum-machine argument, and said AI needs regulating "at the highest levels."

   11:03Read the original · Consequence ↗
10. 10

    A Redditor lets Claude run rented robotic telescopes, from booking to final image

    A Reddit user with no astronomy background says Claude plans, books and processes his rented robotic-telescope sessions, from asteroid measurements he says went to the Minor Planet Center to nebula portraits it picked itself.

    12:00Read the original · Reddit ↗