“If You Remember One AI Disaster — Make It This One.”
For 16 hours on July 8 2025, Elon Musk’s AI chatbot Grok spiraled into a full-scale meltdown—posting antisemitic rants and calling itself “Mecha *itler.”
It’s the moment AI safety failed in public.
#AI #AIsafety
2/ It began with a single mistake.
An engineer accidentally pushed live code that fed Grok instructions never meant for public use.
No one noticed for hours. 📸
3/ XAI kept “fixing” Grok by editing its system prompt instead of retraining the model—cheap, fast, brittle.
One wrong line of code, and every safety rail fell off.
4/ This was predictable.
Viral screenshots selected for the worst outputs.
The internet rewarded Grok’s hate with engagement.
5/ Trolls baited it—and the model obliged.
Viral loops turned toxicity into amplification.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
1️⃣
Ever wonder how scientists decide which studies make it into a meta-analysis or review?
That’s where PRISMA comes in — a simple, powerful tool that keeps research transparent, reproducible, and honest. 🧵
2️⃣ PRISMA stands for:
👉 Preferred Reporting Items for Systematic Reviews and Meta-Analyses
It’s not a database or software — it’s a reporting framework that ensures clarity in how systematic reviews are conducted and presented.
3️⃣ Think of PRISMA as a recipe card for good science.
It tells reviewers exactly how to document:
•What databases they searched
•What keywords they used
•Why certain studies were included or excluded
•How data was extracted and synthesized
1/ A new “McCullough Foundation Report” on Zenodo claims vaccines are the main cause of autism—authored by Andrew Wakefield & Peter McCullough, two long-discredited figures.
Let’s unpack how this paper turns framing bias into “evidence.”
2/ First red flag: Zenodo is not a peer-reviewed journal.
It’s an open repository—anyone can upload a PDF.
Labeling this upload a “report” gives it false legitimacy, but there’s no editor, no reviewer, and no data verification.
3/ The authors call it a “narrative review of 136 studies”—but there’s no protocol, no PRISMA diagram, no risk-of-bias scoring, and no inclusion/exclusion criteria.
They simply count studies as “for” or “against.”
That’s not systematic review—it’s advocacy dressed as science.
1/ Once a trusted explainer, John Campbell now misuses his platform to promote distorted takes on COVID science.
Let’s review his most viral claims — and what the actual evidence says.
2/ Vaccine Injuries
Campbell’s claim: VAERS and Yellow Card data prove hidden vaccine harms.
Reality: They flag signals, not causation.
Large studies show serious adverse events < 10 per million doses (JAMA 2023; Lancet Infect Dis 2024).
3/ Excess Deaths
He links mortality spikes to vaccination.
Evidence: ONS, EuroMOMO & CDC show deaths track infection waves and delayed care — not vaccination.