Joe retweeted
Well shit - grok 4.5 is KILLING it it in my video studio evals!
6/33 -> 23/33 with ridiculous cost efficiency compared to 5.6 models
Joe retweeted
Using Grok agents in background mode
Joe retweeted
NEW: Grok 4.5 now scores highest on the SWE-Atlas-QnA benchmark, edging Claude Fable 5 & GPT-5.6 Sol.
Joe retweeted
Elon Musk explains why the Sun is humanity's ultimate energy source:
"The vast majority of Earth's energy already comes from the Sun. Without it, Earth would be a frozen ice ball
A roughly 100-mile by 100-mile area of solar panels....about a corner of Arizona, could generate enough electricity to power the entire United States"
The challenge isn't whether enough solar energy exists
It's building the infrastructure to capture, store, and distribute it efficiently
Joe retweeted
Grok Build improvements
Grok Build just received another update, bringing smoother navigation and improved keyboard workflows across the Agent Dashboard
Release Notes: v0.2.99 — 2026-07-12
Features:
• Multiline input now works on the agent dashboard the same way it does in regular sessions.
• PageUp and PageDown now scroll the conversation while the prompt is focused.
• Keyboard Shortcuts modal now follows Vim mode navigation keys when enabled
Joe retweeted
The mega bull case for AI infrastructure would be *if* market share shifted away from certain frontier labs with 90%+ inference margins toward cheaper models, whether open-source or closed.
It would increase the ROI on AI spend for end customers by increasing intelligence per dollar, which would drive incremental token demand. Margin dollars would effectively get redistributed from the frontier labs to AI infrastructure providers. The infra winners would be those with the lowest per token cost and the winners at the model layer would be those with the highest token efficiency.
There are many reasons Jensen is so focused on open source, but this is likely the most important one as I think he is probably less worried about a monopsony these days. Lower margin % at the model layer = more margin $ at the infra layer all else equal.
With SpaceX and Meta being vertically integrated and possessing the #3 and #4 models respectively it is more possible than ever. Note that Grok 4.5 is ahead of Fable for some useful tasks at a much lower cost, so ranking them #3 is conservative.
This is not happening yet. Cheap, mostly open source tokens are likely the majority of volume today but the majority of economic value is still accruing to the most intelligent models. Might change though.
We will see.
This is true as I have heard this from contacts in the Valley. Goes with my pinned post.
The AI race is shifting from bigger models to cheaper, smarter systems cnbc.com/2026/07/10/the-ai-r…
Joe retweeted
Follow the Starship series @SpaceX
The path to launch is filled with obstacles and success is only possible through the tireless efforts of many working together towards a common goal. “Critical Path” continues the ongoing Starship series, following SpaceX engineers through the final days before launch of the first Starship V3 and the challenges that come with development of the world’s most powerful and fully reusable rocket.
Joe retweeted
Cool
I made myself a personalized podcast player using @grok 4.5 and @perplexity_ai Computer
It finds the top topics people are discussing on tech and business podcasts, then lets me play a single stream across multiple podcasts by topic.
When you click PLAY DEEP-LINK STREAM, it plays me all the pods talking about China and open source (in the screenshot here)
it's called PODMEME (after Techmeme, which is what i modeled it off of!)
Thanks @AravSrinivas @elonmusk for the harness and model! Cost me 1,100 "credits" on Perplexity -- so $11!
Not sure what it will cost to run this every 48 hours... but, I think worth it.
I might actually output a custom MP4 file for each topic with an audio transition in the middle
Joe retweeted
Grok 4.5 is Opus class for browser use
Joe retweeted
For the first time, Starship will carry 20 next-generation Starlink V3 satellites to space
Starlink V3 is designed to massively expand the network’s capacity and increase user speeds
After deployment, the test satellites will:
• Extend their solar arrays and antennas
• Attempt to connect with ground stations in South Africa
• Establish links with the wider Starlink constellation using high-capacity lasers
The satellites will follow the same suborbital trajectory as Starship, meaning they will not remain in orbit and are expected to reenter the atmosphere later in the mission
But the most interesting part is how Starlink will help inspect Starship
Six of the satellites have been modified with cameras designed to scan Starship’s heat shield and transmit the imagery back to SpaceX operators
Several heat-shield tiles have even been painted white to simulate missing tiles and provide clear imaging targets
The goal is to continue testing methods of analyzing Starship’s heat shield readiness for return to launch site on future missions
Starship launches Starlink
Starlink helps inspect Starship
The entire SpaceX ecosystem is starting to compound on itself
Joe retweeted
I say both energy and mass, as they are separate scarcities for practical purposes, despite being technically “equivalent”, as described by Einstein’s famous equation.
Our solar system has only a tiny percentage of non-solar atoms with which to build things.
Joe retweeted
I would like to offer a counterargument that LLMs (or maybe AIs) cannot jump.
Before AlphaGo, the AI field had the same argument for Go: there are 2.08 × 10^170 possibilities, nothing fits in the computer, and there is no way AI could possibly predict the outcome of the next 50-60 moves.
It turned out most moves do not lead to a win. Combined with clever use of Monte Carlo Tree Search, the sampling becomes quite manageable. The same can be said for physics, where equations are just another form of compression.
Einstein did not start with relativity. That was not his first paper. He spent years understanding the properties of light before concluding that the speed of light is constant across the universe, which unlocked his discovery of relativity. During his thought process, he also interacted with other physicists (e.g., sub-agents) to enrich his thinking.
Currently we have not run an agent for years of compute. The sessions are often fragmented and disoriented, so every new session is almost a fragmented memory of the past, but it may not be for long.
interesting position paper throwing cold water on autoresearch/ai scientist: LLMs can't jump.
The thought experiment is this: Take an LLM with a 1905 knowledge cutoff. Feed it every paper, every dataset, every equation of that era. Could it invent general relativity?
No.
Discovery isn't one thing. It's three. You can induce — generalize from data, which lands you at Newton plus some epicycles to explain Mercury's weird orbit. You can deduce — derive rigorously from axioms you already have, which never gives you new axioms. Or you can jump — invent the frame itself, decide that spacetime curves. That third move is the one that matters, and it's exactly the one induction and deduction can't reach.
Penrose put it as three worlds: Physical, Mental, Platonic. Data flows from the world into a mind fine. But the new law has to be discovered into the Platonic world first — and that step is the jump. LLMs are induction machines running over what already exists. Structurally, they don't take it.
I think it’s a warning to AI scientists/autoresearch against collapsing two very different things into one word.
Hill-climbing: LLMs are already superhuman here, and autoresearch in this sense is real and moving fast.
Abduction/leap/jump: a new frame that reorganizes the field, that is a different act entirely, and nothing about scaling induction suggests you get there.
Most of what Autoresearch ships today will be spectacular hill-climbing. The jump is still ours for now.
Joe retweeted
How many people have left America to escape Capitalism?
VS
How many people have come to America while escaping socialism?
Joe retweeted
Starship is evolving rapidly
SpaceX just announced that they will be deploying V3 Starlink satellites for the first time during their 13th Starship test flight next week! Six of the satellites have been modified with a suite of cameras to scan Starship’s heat shield and transmit imagery.
Summary of every major upgrade and modification @SpaceX says it's made for the upcoming test flight:
• More robust stage separation flip sequence to prevent the booster from rotating off course.
• Hardware upgrades to improve Super Heavy Raptor re-light reliability during the boostback burn.
• Updated engine alarms and abort logic for multi-engine flight conditions.
• Starship propulsion system hardware and operational changes to address the Flight 12 engine-out issue.
• First deployment of 20 Starlink V3 satellites with laser links, deployable solar arrays, and antennas.
• Six Starlink V3 satellites equipped with cameras to inspect Starship's heat shield after reentry.
• White-painted heat shield tiles added as simulated missing tiles and imaging targets.
• New heat shield tile designs and attachment mechanisms, including tests on the aft flaps and aft skirt.
• Load-sensing heat shield tiles to measure stresses during flight.
• Higher dynamic pressure ascent profile to stress-test the heat shield and increase payload capability.
• Planned in-space Raptor engine relight test.
SpaceX: "For the first time, Starship will carry V3 Starlink satellites to space, which aim to greatly expand the network's capacity and user speeds. As part of this initial test, Starship is planned to deploy 20 satellites which will extend solar arrays and antennas and will attempt to connect with ground stations in South Africa and the larger Starlink constellation via high-capacity lasers. Six of the satellites have been modified with a suite of cameras to scan Starship’s heat shield and transmit imagery down to operators to continue testing methods of analyzing Starship’s heat shield readiness for return to launch site on future missions. Several tiles on Starship have been painted white to simulate missing tiles and serve as imaging targets in the test."
