Is Grok 3 the Crowned Champion of AI?: Just hours ago, another advanced large-language model burst onto the scene, smashing existing benchmarks and snagging the top spot on the LM Arena leaderboard. Meet Grok 3, crafted by Elon Musk’s crew, a model that’s as uncensored as it is bold. It’s not just wicked smart—it’s fearless, churning out content that could get you in hot water in many parts of the world. With a deep-thinking mode akin to DeepSeeker 1, rumored text-to-video skills, and a soon-to-launch paid tier called SuperGrok, this AI is making waves.
I’m already shelling out for Twitter Premium Plus to get my hands on Grok 3, which stings a bit. But in today’s video, we’re diving into what makes Grok 3 stand out, how it was trained, and whether it’s truly the king of large-language models. It’s February 18th, 2025, and you’re tuned into The Code Report.
The AI Game of Thrones
Last week, Elon tried to flex on OpenAI with a cheeky buyout offer. No shocker here—the OpenAI board shot it down fast, keeping Sam Altman on his path to turn the nonprofit into a cash machine. The AI world is a brutal battleground, and Mark Zuckerberg took a hit too. Turns out Meta fed their Llama models 82 terabytes of pirated books from the Library Genesis Project, a shady stash of millions of books and paywalled articles. Zuckerbooks pulling a fast one? Yeah, nobody’s clutching their pearls over that.
Training Grok 3: Twitter’s Data Edge
Grok 3’s secret sauce? It’s hooked straight into Twitter’s firehose of data. The xAI team has tuned it for raw truth-seeking, tossing political correctness out the window. That means it’ll whip up stuff other models won’t touch, like:
- Celebrity pics that’d make your grandma blush
- Profanity-packed poems about racial stereotypes
For science, I threw that poem prompt at every major LLM. All blocked it—except Grok 3. The result was so wild I can’t even show it here. Post that in a country without free speech, and you’re toast. Still, Grok 3’s rolling out soon in places like Germany and the UK. Trolls, rejoice.
Performance Check: Does Grok 3 Rule?
Grok 3’s perched atop the LM Arena leaderboard, where humans blindly pit LLMs against each other. It’s also flexing on Gemini, Claw, DeepSeek, and GPT-4 in math, science, and coding. But hold up—some benchmarks are AWOL:
- Codeforces
- ArcAGI
Toss in OpenAI’s o1, and the leaderboard shifts. Benchmarks are cherry-picked anyway, so I ran my own test. Grok 3 nailed valid Svelte5 code in one go and helped me build a game in Godot. It’s holding its own with the big dogs, though the AI race lately is more about slick prompting tricks like DeepResearch than bigger base models.
Inside the Colossus: Grok 3’s Training Ground
Grok 3 was forged at the Colossus supercomputer in Memphis, Tennessee—likely the planet’s beefiest AI rig. Packing over 200,000 NVIDIA H100 GPUs, it guzzles so much juice they’ve hauled in portable diesel generators to keep it humming. That firepower’s gearing up for SuperGrok, slated to hit at $30 a month. That’s a steal next to ChadCPT Pro’s $200 price tag. Me? I’m broke from subscriptions—Claude, Cursor, Gemini, ChadCPT, Copilot, Codium, Midjourney, WatsonX—and my code’s still a mess.
The Verdict: Grok 3’s Edge and Risks
Grok 3’s a beast—uncensored, powerful, and riding Twitter’s data wave. It shines in benchmarks and real-world tasks, but its no-filter vibe could stir trouble in strict-speech zones. The AI war’s heating up, and while Grok 3’s a contender, the focus is shifting to smarter ways to use these models, not just bulking them up. For trolls and coders alike, it’s a shiny new toy—just don’t get caught with the receipts.