Grok Bot keeps getting more capable! 🚀 The latest release is v0.79.3, and @bot has received some interesting upgrades over the past few days: • Flight booking: Can book flights with Delta, American, Alaska, JetBlue, and Frontier, following their websites' booking processes. • Security: Detects hidden instructions targeting AI agents on websites and refuses to follow them. • Computer use: Supports SSH and PostgreSQL commands on its computer. • Email: Gmail and Outlook drafts now support attachments. • Tesla integration: Sensitive vehicle commands always require explicit approval. • Reliability: Improved web fetching, phone calls, 1Password integration, and numerous bug fixes. Grok Bot is evolving into a genuinely useful autonomous agent that can handle real-world tasks, not just answer questions.

There is this saying: "This is the worst AI will ever be." This used to be true, but now it's becoming clear that if the wrong people take over the government, AI could actually get worse. Or become banned altogether. Think about it hard! ⚠️

Elon Musk gave an update on Grok 4.8: "Grok 4.8 will bear no resemblance to 4.7 (thankfully). It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special." It's currently in RL training, initial testing this week.

A robot grabs a toy part, the part slips, and the robot tries again. That sounds trivial until you ask it to assemble the same toy 100 times in a row. @Nvidia ran that test with three different toys. Its GR00T 2 system completed at least 97 assemblies out of 100 for each one, recovering from stalls and mistakes along the way. GR00T 2 predicts what the robot’s cameras should see after its next move. The robot makes the move, looks at what actually happened, and uses that view to plan the next one. So a bad grip can turn into a second attempt instead of ending the run. Nvidia has published the demos and says the code is coming in December.

New trailer for BLADE RUNNER 2099 by Prime Video! Many are worried that it could be too woke. We'll have to see about that. But I think they nailed the visual style! Check it out:

DeepSeek’s open-source agent app is already at 1.54 million daily users, according to the company. @deepseek_ai says Harness users consume around 92 trillion tokens a day, and that only counts people using its official API. The community has already built around 35,000 third-party plugins. The part I find most interesting: plugins can replace built-in features, so developers can change how the agent itself works. DeepSeek is now planning an official marketplace and token grants for plugin authors. If you’re building a plugin, that gives you somewhere to distribute it and some compute to test it with.

Google is stepping up its challenge to Nvidia's dominance in AI hardware. Today, @Google unveiled two new TPU variants designed for specialized AI workloads. The TPU 8tx targets secure AI computing for governments and regulated industries, delivering up to 2x the performance of its predecessor. The TPU 8i Max is designed for demanding workloads like coding agents, mathematical reasoning, and long-context AI applications. Google is also expanding access to its TPUs beyond its own infrastructure, including a partnership with Oracle to bring them to Oracle Cloud in 2027.

Anthropic is reportedly spending $3.4 billion just to make sure its data centers have power. The Financial Times says @AnthropicAI secured commitments for 1.7 gigawatts of natural gas generators from Texas supplier OnePWR in late September. The deal also reportedly includes engineering and construction support to install them. Buying more GPUs only gets you so far if you can’t turn them on. At this scale, getting electricity sooner can mean bringing an entire data center online sooner. That’s what makes this deal so interesting: Anthropic is willing to spend billions to get more control over when that happens.