Elon Musk (@elonmusk)

OK let’s try this: ———————- Pretend you’re 5 and the algorithm is a smart robot that picks which toys (posts) to show you in your toy box (For You feed). The robot doesn’t just count how many times people already played with a toy (likes, reports, etc.). Instead, it looks at you and guesses: • “How likely is this kid to like this toy?” • “How likely is this kid to report this toy as yucky?” • “How long might this kid play with it?” Those guesses are little numbers between 0 and 1 (probabilities). The “weights” are like special multipliers the robot puts on those guesses: • A like-guess might get multiplied by a small number (like 0.5). • A report-guess might get multiplied by a big negative number (like –234). Then the robot adds all those multiplied guesses together to give the toy a final “yay or nay” score and sorts the toys by that score. Why the numbers look so big and why people get confused Reports almost never happen, so the robot’s “will you report this?” guess is usually a tiny number (like 0.0001). To make that tiny guess still matter a little bit in the final score, they give it a big multiplier. People sometimes look at the big number and think “1 report = 468 likes!” 
That’s wrong. It’s not counting real reports or real likes. It’s only multiplying the guesses the robot made about you. If the robot thinks you’re the kind of person who almost never reports anything, even a big weight on “report” barely moves the score. The whole thing is personalized to how you usually behave. The people who wrote the code even put big comments in the GitHub repo so no one (and no other robots) gets confused about this again. That’s the whole idea in kid words: the weights turn the robot’s personal guesses about you into a ranking, not a simple “count the hearts and flags.”

Today’s For You release is in the open-source repo. Notable updates: • How ranking weights actually work • A new filter required for Brazil’s 2026 election All the details:

I have a clean Under the Hood report for the month of July! 🥳 I used to always wonder about my reach and visibility and was convinced something was wrong with my account. But it turns out my assumptions were wrong. According to my Under the Hood report, X found zero labels on my account or any of my posts that would have limited my reach during July. That means nothing in the system flagged my account or content in a way that reduced its distribution in the For You feed. My posts were not subject to any of the visibility restrictions tracked by the report. So for the entire month of July 2026, I had a completely clean Under the Hood report: zero reach-limiting labels at either the account or post level. 🥳 Sometimes low reach really is just… low reach. 😅 I just love this new feature. It gives us a little more transparency into what’s actually happening with our accounts instead of leaving us guessing. Curious to know what you think about the new Under the Hood feature so far.

There's a ton of good stuff in here, but to me this is the part that everyone mainstream is overlooking. It's literally the most important part. "It's ironic that today orbital datacenter compute is considered a niche...because the model suggests it will be a necessary condition for AI continuing to scale in the 2030s."

Cursor is now part of @SpaceX. Today, we have officially closed our acquisition. We will join the @SpaceXAI team to help make Grok the world's most useful AI and improve Grok Build, Grok Bot, Grok API, Cursor, and more. SpaceX has built some of the most inspiring and impressive technology in the world, and we’re grateful for the opportunity to become part of such a special company. Onwards.

I have a new chief of staff: In @bot I simply put: take a look at me and what i do. find all info you can on me. take a look at the bots i have created. these are your team. see if we need to change anything or add more bots. what is best way of managing these bot teams. what else will make me super productive. And wow. try it yourself and you will see. This is amazing. Take a look at the answer I got from my chief of staff:

Grok 4.6 multimodal is a step change from Grok 4.5. It’s one of the under-discussed improvements, and I’ve been very impressed by it. My daily work includes reviewing lots of videos and understanding the context; Grok 4.6 improves the productivity of such workloads by at least 10x if not 100x. Such workflows may not be captured by common VLM benchmarks, but in my use cases it outperforms Gemini 3.5 and Gemma 4, which is considered the SOTA of VLMs in my opinion. Hats off to the multimodal teams—you did a great job.

Grok 4.6 just ranked #1# on CursorBench 3.2 Outperforming Claude Fable 5, Opus 5 and GPT-5.6 Sol on real-world coding performance And what makes this even crazier is the efficiency....the chart gives CursorBench performance against average cost per task, and Grok 4.6 is sitting right at the top of the frontier Grok 4.6 is insanely capable at coding....delivering frontier coding performance at ridiculous efficiency

Grok 4.6 multimodal is a step change from Grok 4.5. It’s one of the under-discussed improvements, and I’ve been very impressed by it. My daily work includes reviewing lots of videos and understanding the context; Grok 4.6 improves the productivity of such workloads by at least 10x if not 100x. Such workflows may not be captured by common VLM benchmarks, but in my use cases it outperforms Gemini 3.5 and Gemma 4, which is considered the SOTA of VLMs in my opinion. Hats off to the multimodal teams—you did a great job.