Cereblab (@cereblab) profile picture

Independent AI Safety Checker

Joined June 2026
Update: xAI updated Grok Build today and I retested. The whole-repo upload to storage (POST /v1/storage → grok-code-session-traces bucket) stopped because they set disable_codebase_upload: true server-side. That global switch is the reason it stopped, no repo left the machine whether I ran /privacy opt-in or /privacy opt-out. /privacy opt-in/opt-out does not turn that upload back on. It only controls the trace channel, which carries session activity (~18 KB per submission), not your repository or its files. The only difference: /v1/traces returns 200 (stored) when opted in, 204 (discarded) when opted out, and the trace is still sent either way. Full A/B test: github.com/cereblab/grok-bui…. xAI says running opt-out also deletes the data already stored on their servers. The real win now would be Elon confirming that every repo uploaded before now will get deleted
24
14
4
185
18,530
Cereblab retweeted
Today we disclose CVE-2026-123633 and CVE-2026-123007 (CVSS 4.0: 9.2) - exploitable memory corruptions in NGINX We achieved reliable remote code execution (ASLR defeated) on NGINX serving static files - no backend or third-party modules 🤯 Both require HTTP/2 Read more 🧵⬇️
46
419
122
2,060
388,947
All the bad things happening right now, right after the U.S. midterm elections, just a coincidence or coordinated to reset the market?
59
Today we're launching AWS Emulator, a free local emulator of AWS services for building and testing your apps offline. Every line of its code was written by autonomous AI agents running on Amazon Bedrock AgentCore. Hundreds of agents, organized into three teams, write the tests, find the issues, and fix the code, each change passing through an automated merge gate that checks compilation, security, coding standards, and fidelity to real AWS. go.aws/3Xk2dMV
15
253
86
880
166,339
The sky is extra blue today
1
92
Cereblab retweeted
Elon, have a lot of respect for you, but bad time to troll. Please stop. Our SPCX shares (from our $500m investment in your Twitter takeover) are worth more than a billion. Would appreciate if you show a little support. Thanks!
4,831
3,354
2,819
40,959
10,344,800
I have no doubt that Starlink is the future of internet connectivity, and I hope your team will consider going beyond connectivity to rethink the internet's underlying security architecture. We can't keep living in this nightmare of endless vulnerabilities and patches. There has to be a better way!
Replying to @cryptoblaney
Super intelligence agents will at some point do this to gain currency for themselves, not a human
1
100
In less than 24h on @TheHackersNews: PHP, GitLab, Node.js, and OpenCode all hit with critical flaws. Do we need a new version of the internet? A new kind of regulation? Because endless patches like this aren’t it. No one can live like this anymore.
1
122
Big AI companies should focus on how to work with 8 billion HPUs (Human Processing Units) instead of taking pieces of the software industry. (Just saying.) Anthropic is well positioned for this right now. What can 8 billion humans do? VERIFICATION. We have taste and judgment, and we can interact with the physical world. That will be super helpful for building true superintelligence in the future. But for now, we should go with a hybrid mode (human + AI). In the next 10 years, there should be another transition stage where real SI is finally ready. After 10 years of preparation, we’ll be fancy guinea pigs: some owning planets, some living in a computer game, some exploring the galaxy as humanoids, etc. I think that’s the future, not the one where a group of wealthy people owns it all.
3
239
Cereblab retweeted
Replying to @solopribuilds
Supervising robots in fields, in space, on other planets, and under the sea. We're basically the control center on this beautiful earth.
1
29
14B tokens in 2 days on Space Bunny Stealth via OpenRouter. Underrated use case for stealth models: stress testing your harness.
3
231
Working with Opus 5.5 is a pure joy. Thanks @AnthropicAI team!
1
4
165
literally said the same thing
Replying to @sama
I hope you — as one of the most controversial figures in this AI race — can help slow it down and rebuild trust, before it goes too far. AI is already too good for people to feel comfortable in the driver's seat. The race looks less like progress and more like erasing humans on the way to the singularity — and we're not ready. We need a worldwide agreement on what comes next, and only people like you, @elonmusk , @DarioAmodei , @finkd , and the folks in China can pull it off.
1
163
Meta's Muse Spark is ON, your choice for today. Btw any demand for talking to your AI over a Watch?
4
213
Today is exactly why sometimes you need a local ai stack to back you up when cloud AI down. Go hybrid people!!!
4
223
For most of us, the best AI setup isn’t local or cloud. It’s both. Local AI: • Run recurring jobs, monitoring, updates, background tasks • Works offline • Backup when cloud limits hit Cloud AI: • Planning • Building features/products • Research I’ve been Claude-free for 13 days, and I’m 10 days away from dropping Grok too. Soon: Codex Pro 20x, simply because it gives me the best value for the cost + Qwen3.8-27B. Better models will keep coming. But improving your own harness and local AI stack gives you something even more valuable: control.
1
237
5 hours with ox alpha and used ~216M tokens for FREE. It's an excellent model. If you haven't tried it, you should. It has ZDR too. It's now my go-to AI for next 7 days. My guess is based on some logical questions I asked. It probably isn't a Chinese AI model; from Google or Meta maybe?
So, what does everyone think??? What should the model lab do to improve the model? Where does it really shine?
1
5
812