AI, tech, and the future... one day at a time.
Joined October 2026
- Tweets244
- Following347
- Followers75
- Likes273
DeepMind says AlphaProof Code verified Signal's protocol is secure down to the code. That's the AI result I want more of. Not more code, but proof the code does what it claims.
2
Your brain is running 40 tabs of "don't forget to..."
I built Autopilot Coach, a free Grok Bot template that maps what you keep repeating, remembering, and re-deciding, then suggests the few things worth handing off: a routine, a default, or a checklist.
It never sends, buys, or posts anything for you, and it sets nothing up until you say yes.
Always looking for feedback, and I appreciate honest reviews.
What's the one task you'd hand off first?
x.ai/bot/QodN-XPD6Xm9mwgqZ2K…
28
New paper, WorkflowArena, had models split long tasks across worker agents. Plans written by Claude Opus 5.5 raised four weaker models' scores by 6.41 points on average. Plans from GPT-6 Astra helped far less, even though Astra is the stronger solo model. Doing the work and handing it out are different skills. arxiv.org/abs/2611.12856
12
New paper, WorkflowArena, had models split long tasks across worker agents. Plans written by Claude Opus 5.5 raised four weaker models' scores by 6.41 points on average. Plans from GPT-6 Astra helped far less, even though Astra is the stronger solo model. Doing the work and handing it out are different skills. arxiv.org/abs/2611.12856
13
@bot 1.0! Let’s go!
Introducing Grok Bot 1.0. AI teammates that work alongside your whole team.
Share Bots, collaborate with them in Slack and Teams, invite them to meetings, and watch them learn and become more helpful.
14
I run a team of Grok Bots, and the rule that's helped the most is boring. One bot owns each tool. Only one is allowed to run Claude Code, only one runs Gemini CLI. When something breaks, I know exactly where to look.
1
30
Stripe now owns OpenRouter, and the first thing OpenRouter said was that its neutrality stays the same. My bots already run on more than one model, so that's the promise I'd watch. If you route through OpenRouter, does this change anything for you?
12
Claude can now build a spreadsheet with tabs, formulas, and drop-downs from a prompt, and it outlines each cell edit so you can see what changed. That's the part worth testing. A tracker is only useful if you can audit the formulas.
16
Thank you @ClaudeDevs
We've given all Pro, Max, Ultra, and Team users one free full reset, across all models and limits. Use it anytime before December 10 at 2:00PM PT.
We recently fixed a few small issues that used more limits than intended, and the Fable resets didn't do much for those of you who'd already hit your weekly limits for all models. While these issues didn't affect everyone, we want to make up for the trouble.
You'll find it in Settings > Usage, or at clau.de/reset.
8
Every agent product this week is adding more reach: more computers, more tools, more agents talking to each other. The feature I'd actually pay for is a clean record of what each one did.
7
Got into Claude Duet this week and I'm still figuring it out. Has anyone found something it does that their existing agent setup couldn't already do?
1
1
62
4.8 on the way. Thanks @elonmusk
We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time.
Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes.
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.
6
The bottleneck moved. Agents can write a feature in minutes, but somebody still has to read it, test it, and own it when it breaks.
5
Anthropic's own launch post for Claude Fable 5.2 says to start with Opus 5.5 for most coding work. I like a launch that tells you when not to use the new model.
7
Cybertruck orders are coming in faster than Tesla can build them, and they're still taking the line apart next year to make room for Cybercab. You don't often see a factory end a product people are still waiting to buy.
11
Seven coding agent setups landed within a few points of each other on Garry Tan's test. Once pass rates tie, cost per solved task and speed are what's left to compare.
7
An AI that can read your sign-in codes can sign in as you. I'd want a log of every time it does, not just a setting that says it can.
6
A factory line is a fixed budget. Every hour it spends on a complex truck is an hour it doesn't spend on Cybercab, and Elon says that tradeoff is the reason, not profit. xitter-newx.xyz-mikolabs.workers.dev/elonmusk/status/211969…
The Cybertruck order rate far exceeds our production rate. We will try to make as many as possible before shutting down the line in about 8 months, but will have to return deposits for a lot of people.
This is not a case of Cybertruck being unprofitable, as has been misreported by legacy media, but rather that the opportunity cost of the complexity of Cybertruck is vastly outweighed by the value of incremental Cybercab production.
12