Musk's xAI Releases Grok-4: How Far Can an AI That Champions 'Anti-Political Correctness' Go?

Musk's xAI company has released Grok-4.

Positioning

Grok-4's core selling point is "no censorship" — Musk said at the launch event: "Other AIs are trained to be overly politically correct. Grok will tell you what it really thinks."

Specifically, Grok's responses on sensitive topics (politics, religion, gender) are indeed more direct than ChatGPT's, and it won't constantly fall back on lines like "As an AI assistant, I cannot comment on this issue."

Technical Capabilities

• Parameter size: undisclosed (rumored to be ~800B MoE)
• Training data: real-time Twitter/X data is a unique advantage
• Reasoning ability: close to GPT-4o in some benchmarks

Controversy

• Some people feel Grok's "no censorship" is actually a new form of bias — shifting from leaning left to leaning right
• EU data regulators have already begun investigating whether Grok violates the DSA (Digital Services Act)
• Advertisers are cautious — they're hesitant to place ads next to an "uncensored" AI

My Take

Grok's market positioning is smart — while ChatGPT and Claude become increasingly "safe," there will always be people who want an AI with fewer constraints. But "no censorship" is a double-edged sword that can easily cross the line.

Musk's playbook is always to maximize buzz first and figure out the rest later.