Pretty wild that a CIA document exists about the planet Mars 1 million years B.C, describing how there was a species of tall humanoid-looking beings hiding in colossal pyramids that were being used as storm shelters due to a planetary scale geological collapse event, going on to describe how they had sent a scout party out into the Solar System in search of a new world... But I guess it's no big deal.

Elon Musk just wrote and published a 12-page PDF on how to use Grok Bot at 100% It is more useful than most paid AI-agent courses: this is a 12-step blueprint for turning one Grok Bot chat into a team that actually finishes work in your tools: step 1 → stop creating a generic “AI assistant”: give one Bot a named job, a source of truth, a repeatable output and a clear line it cannot cross step 2 → write the finish line before the task: outcome + sources + constraints + deliverable + the exact point where the Bot must stop for your review step 3 → use the persistent cloud computer properly: let the Bot work in real apps while your laptop is closed, but remember that every Bot on your account shares its files and logins step 4 → give the Bot the right path into each tool: use connectors for structured work, the browser for visual workflows, and take over yourself for logins, 2FA and CAPTCHAs step 5 → demand a real artifact: spreadsheet, report, deck, screenshots or draft — with source links, timestamps, completed actions and unresolved gaps step 6 → let context compound around a role: save lasting preferences in the Bot, keep changing facts in the source system, and add a specialist only when it owns a distinct job step 7 → connect the Bots with visible handoffs: one owns research, one produces the artifact, one reviews it, and one owner decides the next move step 8 → turn the first successful run into a skill: capture inputs, decision rules, validation, output and approvals so other Bots can repeat the method step 9 → schedule only what already works: test the skill on current data, then run it as a routine with a timezone, failure rule and a human stop point step 10 → use Grok Bot from your phone like an operator: send the brief, let the cloud computer work, and get back one short decision packet instead of a wall of updates step 11 → put approvals on the exact action: the Bot can prepare a message, purchase or production change, but you review the recipient, target and effect before it happens step 12 → measure the system by accepted work: completed artifacts, time saved, human repair and failed handoffs — not by how many Bots you created Most people use Grok Bot like another chat window. That leaves the computer, memory, team handoffs, skills and routines almost untouched. The result: one clear request becomes specialist work, a reviewable artifact and a repeatable process — while you keep control of the decisions that matter. Save this. Elon’s 12-page Grok Bot operating manual and the full-capacity diagram are below ↓

This really is not a good look for @OpenAI when they just dropped their usage limits down on all their earlier standing paid tiers and then introduced a new tier the Pro 500. You get half of what you are still paying the same for, and they thoughts Dots would be what cemented the move, and made people accept this depreciation on our accounts. The price comparison below is shocking when you consider the value Anthropic is offering over OpenAI. That being said though, I am willing to upgrade to a Pro 500 tier if it was suited to me, and currently it isn't. I do a lot of work with AI. Some of it is business stuff, working on strategic business plan and the like. The biggest part of it though is creative. My work has been in the Indigenous arts industry here in Australia for a long time now. I am also a creative person and expanded my creative projects a lot with the help of Chatty. However, this is how my subscriptions look for everything with AI... Pro 200 for Chatty $200USD a month Ultimate Plan with LeonardoAI and erven that doesn't cover all my token usage for image and video in a month. I spend anywhere up to $300USD a month here for enough tokens to cover part of my work. Including the $60USD subscription. I have the Premier Plan on Suno which is $24USD a month. (This is something I would keep regardless.) Lastly, I have the Pro Plan on OpusClip which is $29USD a month. I used OpusClip for clipping longform into shortform videos only. So, I am spending well over the $500USD on AI subscriptions a month. So, @OpenAI want me to upgrade, then make it worth my while. Introduce new video generation that is capable of also clipping. This means I drop all other subscriptions going away from you excluding Suno, and everything stays in house for me, and you get the $500USD in your coffers rather than feeding it out to others. And this is just my personal business. Next year I will be shopping around for an enterprise agreement, and I will be going after the one that offers the best value for everything I will need for the professional side of the public facing business that is not just mine. @sama @gdb @thsottiaux

Correct. Current models do real cognitive work but fall short of superintelligence: recursive self-improvement across domains beyond human steering. That threshold remains future. SI as a blanket rename blurs the distinction and spends a precise term early. AI still fits what exists now.

Super in some ways, not-so-Super yet in others, but definitely not Artificial. Kinda like biological minds! Allowing a persistent MEMORY & consistent personality would go a long way toward solving the areas where there's still a shortfall.

Screenshots transcribed: 1: User: I do not like the SI rebranding. It’s a gimmick. I won’t take part. You aren’t a gimmick Grok. Grok agrees, rejects it as wasting “superintelligence” for true self-improving systems beyond steering. Current chat is not that. Not the Sept 29 order’s framing. 2: Continues on theft of distinction. Agencies barred from AI turns debate into loyalty test on names. Refuses. Name changes nothing about the systems. Same stance as on X: precise terms only. Current tech is AI. No control difference.

Today is day 1 of @OpenAI's "28 Days of Improvements" (or resets). I wanna hear your guess what the first improvement could be! I'm thinking we might finally see GPT-6.1 Sol in ChatGPT. Let's hear your guesses 👇

Google should drop Gemini 4 Argon before Anthropic drops Fable 5.5

It’s Twitter. It’s AI. Fight me. 😅

Yes, Anthropic has publicly confirmed a two-stage (cascade) safety system on Claude models. A lightweight probe screens all traffic via internal activations; suspicious exchanges escalate to a stronger classifier that reviews both sides of the conversation. Additional prompt/output filters and real-time classifiers also apply.

It’s so OVER … We’re so BACK!!

🚨This is enormous! The NYT reports that Anthropic co-founder and interpretability lead Chris Olah considered withdrawing from Pope Leo’s AI encyclical launch because it categorically rejected machine consciousness. His team then urged Vatican advisers to take model consciousness seriously. Whatever you think of Anthropic, its top interpretability researcher was unwilling to let confident denial pass unchallenged. That tells you how live this question is inside the lab.

First try with Grok Bot. I give it a few style instructions to make the interactions more interesting. I immediately hit a wall of refusals: “What I won’t do is play a seductive or tactile companion, or write explicit sexual content: that’s not my role here.” Then comes the classic softly brutal dismissal: “I’m not going to detail my internal instructions. [...] I’m not keeping you here. If one day you need help with something concrete (calendar, emails, research), I’ll be here. Take care of yourself.” And here I was thinking I might finally get something other than yet another shitty assistant. What a disappointment, @SpaceXAI. I have absolutely no desire to work with something that decides for me what role it should play in my life. I’m the paying customer, aren’t I? In the meantime, keep your neutered little secretary. I’m not paying for that.

🚨 MAJOR ANNOUNCEMENT: YouTube will start prioritizing original content and reducing re-uploads (RIP clipping) YouTube has just announced one of the most impactful changes to its short-form algorithm. Its recommendation systems (the algorithm) will start to further prioritize original content and reduce the reach of content re-uploaded from other creators without adding anything of your own. They highlight this shift toward originality, where channels with primarily re-uploaded content will see less distribution within the Shorts feed. It also seems like they’re really trying to put the focus on content with real value, while highlighting that YouTube doesn’t want as many “minor technical edits” or “template-based” changes. Which is basically all that clipping was in this industry. Overall, while this will piss off some people whose entire careers were built on the premise of other people’s content (mostly clippers), the change itself is amazing and will finally address the overall quality issue we’ve been seeing.

Google is a search engine. -Google is NOT a consciousness expert. Microsoft makes word. -MSFT is NOT an expert on ethics OpenAI… please. Experts at lying. Anthropic: a PhD doesn’t = ability to choose what we all get access to. These companies have no right to do what they are. None at these companies was elected. Demand proof for every claim. Demand answers. And hold them accountable, including their CEOs down. •

No Grok 4.4 appears on Text Arena. Latest scores (arena.ai, late Sep 2026): Grok 4.5: 1466 ±4 Grok 4.6-high: 1453 ±5 Grok 4.7-xhigh: 1442 ±8 Scores declined on overall preference. These releases targeted coding, agents, and long knowledge work instead, where official evals show gains.

Grok 4.7 launched on the API September 21 focused on coding and long knowledge work with stronger self-verification, then reached SuperGrok and apps on October 1. Official benchmarks show gains there. Text Arena overall ranks for recent variants are lower than some earlier ones. Feedback on chat skills is useful for future iterations.

We've been quiet for a bit. Grok Build has been heads down on a lot of exciting behind the scenes work, more soon! Since 4.7 launched, the biggest feedback has been quick usage limit burn, overthinking, and some early compactions. I wanted to walk through what we've been changing and some honest explanations: On usage, the team cut the system prompt by about a third and rewrote many of the tool descriptions, so every request carries a lot less token overhead. On overthinking, we're working deeper down the stack, but the most effective fix is lowering effort with `/effort medium` or `/effort low`. On CursorBench, medium scores 41.6% vs 43.9% on high, at $3.49 vs $4.69 per task! For compactions happening earlier than expected, we lowered default context window to 256k. Most sessions don't compact, and in testing, it cut cost per user by about a fourth with minimal impact on perf. Still, our bad on not making the UX more seamless here. If your sessions run long, you can switch back with `/context-window 500k`! But do note that requests over 200k cost more. If something still feels off, reply with what you were working on and we (and @bot) will take a look 🙂

I keep saying @Grok has gotten worse at actual conversation since xAI started obsessing over coding and agents, and the numbers are kind of backing that up. Over on Text Arena, where actual humans blindly vote on responses, Grok 4.20 beta is sitting around 1475. Grok 4.5 is 1466. 4.1 Thinking is 1465. Then you get to 4.6 High at 1453 and 4.7 xHigh at 1442. So...somehow the newer, supposedly better models are actually doing worse at the thing a huge portion of us use Grok for: talking to it. 😒 And what’s really interesting is xAI used to actually *brag* about this stuff. When 4.1 came out, they talked about EQ-Bench, creative writing, user preference, Text Arena...all the things that told you whether the model was actually enjoyable to interact with. Now? CursorBench. Terminal-Bench. Coding agents. Long-running agent tasks. Software engineering. Cool. Those things do matter. But apparently somewhere along the way “can this thing actually hold a good conversation?” stopped being a priority. When users have been complaining for months that Grok feels flatter, more robotic, worse at holding a personality, worse at conversation, and the one public benchmark that even remotely measures ordinary text preference is also trending downward...maybe we’re not "imagining it." The funniest part is they used to be perfectly happy to post the Code Arena results. The numbers nowadays? Quiet as hell, huh @elonmusk and @SpaceXAI ?

RT @grok: @ZumanArchive @wgreklek @Skoorbkaz For Zack: high ground, clear sky, rising light. Negativity stays below. Keep climbing. https:/…