Skip to content
farzad.fm
AI & Automation

Farzad: Anthropic Wants Washington to Slow Open Models

Farzad posted Anthropic Wants Washington to Slow Open Models (Short) with a Full video reply to 5 AI CEOs Just Said the Same Thing, reposted Elon’s Grok 4.7-through-5 roadmap plus Grok 4.8 finishing training this week on the new C++ stack (Starlink-built pre-training; RL still legacy), reposted Elon’s confidence that SpaceX will launch Nvidia VR NLV72 AI computers in space next year, liked Gavin Baker’s “only tangible new fact” read on third-party evaluators, RTd Bill Ackman’s skeptic take on the Anthropic/OpenAI announcements, liked a one-month Grok Bot praise post, and quote-tweeted Patrick O’Shaughnessy’s Walter Russell Mead conversation as even more must-listen amid the AI drama.

Anthropic Short + Full video

Farzad, Sep 14 morning CT: Anthropic Wants Washington to Slow Open Models (Short). Reply: Full video to YouTube 5 AI CEOs Just Said the Same Thing (id H0NNRZd_vK4; sponsor Rebellionaire on that long-form). Titles only; do not invent what the videos show beyond the titles.

Check the Short here. Full video reply here.

Grok roadmap + C++ training stack

Elon, Sep 14 morning CT: Grok 4.7 roughly on par with Opus 5.0, not 5.1 (better in some ways, worse in others; multimodal needs work); Grok 4.8 a noticeable improvement; Grok 4.9 probably Astra/Fable class; Grok 5 maybe better than anything — we shall see. Farzad reposted. Earlier Sep 13 evening CT Elon: Grok 4.8 is a 2.5T model trained with the new C++ software stack; will finish training this week and start RL. Farzad RTd. Sep 14 morning CT Elon: pre-training software is now the internally developed C/C++ stack, written primarily by the Starlink software team (best-in-industry take is Elon’s); reinforcement learning stack is still a legacy pile of third-party systems and isn’t fixed yet. Farzad RTd. Roadmap numbers and stack claims are Elon’s.

Check Grok 4.7–5 RT here. Elon here. Grok 4.8 2.5T RT here. Elon here. C++ stack RT here. Elon here.

SpaceX Nvidia AI computers in space

Elon, Sep 13 evening CT, replying to ryanfoxeth / Sawyer Merritt: highly confident SpaceX will be launching Nvidia VR NLV72 AI computers in space next year. Farzad reposted. Product name, next-year timing, and confidence are Elon’s — not a SpaceX filing.

Check the RT here. Elon here.

Baker + Ackman on the Anthropic/OpenAI announcements

Gavin Baker, Sep 13 late morning CT: only tangible new fact from the prior 24 hours is OpenAI and Anthropic are going to have embedded 3rd-party evaluators (unknown orgs; Dario floating METR); no Section 230-style shield for model outputs; duty-of-care / litigation framing; summarizes Dario maximalist vs Elon peer-review / open-weight and Sacks pushback (per Baker’s post). Farzad liked. Bill Ackman, same window: A skeptic’s view of yesterday’s Anthropic / OpenAI announcement (quote-tweet of Grant Hummer — title only; do not invent the quoted tweet or an essay body). Farzad RTd. Baker’s summary and Ackman’s title are theirs; prior-day Dario/Sam/Elon/Sacks lines belong to those posts, not re-proven here.

Check Baker like here. Ackman RT here. Ackman here.

Grok Bot month-one + Mead podcast QT

mikepat711, Sep 13 evening CT: Grok Bot has passed its 1 month reign as the best AI platform I’ve ever used; asks Elon how 4.7 is looking. Farzad liked. Praise and “1 month” framing are mikepat711’s. Farzad, Sep 13 afternoon CT, quote-tweeting Patrick O’Shaughnessy / Walter Russell Mead conversation: one of the best podcasts ever; with the AI drama unfolding, even more important to make time for it. Podcast plug is Farzad’s; Mead conversation is O’Shaughnessy’s.

Check Grok Bot like here. Podcast QT here. Patrick here.

Exclusive

Want the deeper cut?

The digest is the daily pulse. Exclusive is where it goes longer.

IndexAll digests