Summary
This is the thinnest AI slot of the week. Thirty-eight tweets, zero curated retweets, and exactly one item with real signal: Elon Musk pre-announcing Grok 4.6 in two weeks and Grok 4.7 in four weeks, two weeks after Grok 4.5 shipped. A point release every fortnight is a cadence claim, not a capability claim, and it is worth holding against the fact that xAI just open-sourced its CLI harness. If the model changes every fourteen days, the harness is the only stable surface a developer can build against, which is a reasonable reading of why xAI gave it away. Two carryovers from the morning pull recirculated unchanged: the Grok CLI announcement and the Opus 5 demo thread. Everything else is a wall of geopolitics and Elon fandom that the AI keyword filter let through, with @MarioNawfal contributing eighteen posts on Iran, Hormuz, Houthi escalation and European wildfires, and @brivael contributing fifteen posts of French politics and Musk appreciation, of which exactly one touches compute.
Posts
Grok 4.6 in two weeks, Grok 4.7 in four (@minchoi, quoting @elonmusk). Grok 4.5 landed two weeks ago and xAI is now publicly committing to two more point releases inside a month. The framing in the repost is that the model race has gotten ridiculous, but the operationally interesting part is what a fortnightly release cadence does to anyone evaluating models: benchmark numbers age out before a procurement cycle finishes, and version-pinning becomes the default posture rather than a precaution.
xAI's open-source Grok CLI, recirculating (@aksheyd, quoting @elonmusk, x.ai/cli). Same post as the morning slot, unchanged: a
/tutorialentry point, open source, and a screenshot showing a 500K token counter, session hooks, slash commands, permission modes, and a Grok 4.5 effort selector. Reads differently next to the release-cadence post above. Shipping a free, forkable harness makes sense precisely when the model underneath it is going to be replaced every two weeks.Opus 5 demo thread, recirculating (@minchoi). The same ten-example vibe-coding thread from the morning pull, framed as things people built in the first twenty-four hours. Still demos, still nothing measurable. See Claude Opus 5.
Energy as the real ceiling, via Kardashev (@brivael). The argument is that everything has to scale by 100x, energy first, then compute, and that SpaceX announcing datacenters in space and on the Moon is a physics decision rather than a branding one: continuous solar, no night, no cloud cover, no cooling water. It is written as Musk boosterism and should be discounted as such, but the underlying premise is the same one the energy-to-token position paper made in May, that delivered power and cooling become the binding constraint on inference before peak compute does. Orbital datacenters are a strange answer to that constraint. The constraint itself is real.
Post-scarcity money take (@brivael). Argues that money as we know it becomes obsolete once robots and AI make energy and physical labor nearly free, and that humans will then manufacture a new voluntary scarcity around attention, taste, and meaning. Well written, entirely unfalsifiable, and resting on a labor-cost assumption nobody has demonstrated. Read it as a mood, not a forecast.
Grok CLI livestream and an X article link (@brivael). A French-language biohacking livestream link. Click through to read.
Geopolitics and off-topic bulk (cluster of roughly 32). @MarioNawfal contributed eighteen posts on Iran pausing retaliatory strikes, IRGC warning shots at six ships in the Strait of Hormuz, Houthi threats against Saudi Aramco facilities, US interceptor stockpile depletion from a CSIS report, Russian drone volumes against Ukraine, Italian and Iberian wildfires, a Berlin attack, and assorted entertainment clips. @brivael contributed thirteen more of French domestic politics and Musk appreciation quote-posts. @AustinJustice posted on Flock surveillance cameras and @NICKIMINAJ on a White House Correspondents' Dinner joke. No AI research or industry substance. Skip.