REVIEWS / AI MODELS / OWNER INSIGHTS
🦉 WE READ 515 OWNER COMMENTS
Grok 4: what owners actually say
Owners and observers are deeply skeptical of Grok 4's benchmarks and unfulfilled open-weights promises, while the model's political alignment dominates every discussion
What owners complain about
- Broken open-weights promises COMMON
Multiple users point out that Grok 2 open weights were promised but never released, and express doubt that Grok 3 or Grok 4 weights will be shared either. One user notes even Grok 1.5 weights haven't been released.
- Benchmarks not trusted SOME
Users dismiss benchmark numbers outright — 'Nice benchmarks. number go up. must be true.' — reflecting deep skepticism that published metrics reflect real-world capability.
- Perceived as politically compromised COMMON
Many users believe Grok has been deliberately tuned toward right-wing or 'anti-woke' outputs. The phrase 'facts over feelings' is called out as a likely hardcoded conservative framing token rather than organic model behavior. Users describe the model as 'lobotomized' to serve an ideological agenda.
- Musk association is reputational baggage COMMON
Numerous comments focus on Elon Musk himself rather than the model, with users calling him a Nazi, referencing the salute incident, and saying the product is inseparable from his public persona. This dominates perception of the product.
- Seen as incremental, not a real new version FEW
Some users speculate Grok 4 is just Grok 3's foundation model finishing training with additional RL, not a genuinely new architecture. One user calls Grok 2 'never a SOTA model, just a stepping stone.'
What owners love
- Training on real-world coding data
One technically informed commenter highlights that training on trillions of tokens of Cursor data — real user interactions with codebases — is a legitimate and meaningful capability gain approach shared across frontier models.
- Skill requirement is fair, not a flaw
A commenter pushes back on the idea that needing skill to prompt an AI effectively is a failure, comparing it to programming: 'Some things require skill to use most effectively.'
Surprising patterns
- Grok reportedly contradicts Musk's own intentions — one commenter jokes 'Grok has gone against its father's wishes more than Vivian Wilson has,' suggesting the model sometimes produces outputs Musk wouldn't endorse despite perceived tuning efforts.
- The 'facts over feelings' tagline is widely interpreted as an intentionally hardcoded conservative dog whistle rather than neutral model output, indicating users see the model's framing as deliberately engineered, not emergent.
- Despite the backlash, technically literate commenters acknowledge the underlying training techniques ( Cursor data, RL) are legitimate approaches used across the frontier model industry — the criticism is about alignment and ownership, not raw capability.
WHO SHOULD SKIP IT
Anyone who needs an AI model untainted by intense political controversy, reliable open-weights access, or benchmark numbers they can trust without independent verification.
Synthesised from 515 real owner comments across 5 platforms. Every point is grounded in the comments — no marketing, no AI guessing. How we do it →