Niki Georgiev

Claude vs Codex, building my AI trading lab in public (paper money for now). Tesla FSD daily, solar + Powerwall home. Real numbers, no hype. Florida.

[Niki Georgiev @DrivenByNikiG]

JUST ARRIVED: THE BRAND NEW M5 MAX MACBOOK PRO.

16 inch, came yesterday.
One of the first ones out the door.

What's inside:
M5 Max chip, 18-core CPU, 40-core GPU
64GB of memory
2TB of storage
New OLED touchscreen, up to 2000 nits bright.

And this is only part one.
An M5 Ultra Mac Studio comes in a couple of weeks.

Both are for building my AI trading system, with Claude and Codex running every day.

Let's see what they can do!!!

[Niki Georgiev @DrivenByNikiG]

@davelee That’s amazing. Florida scale please.

[Niki Georgiev @DrivenByNikiG]

@pcshipp A genuine question - how do you use $20 plan? I have a feeling it’ll disappear in less than a day.

[Niki Georgiev @DrivenByNikiG]

Zerohedge: "Elon please hire Jensen."
Elon: "That would be an honor."

Five words, on the same night Nvidia's board reportedly met to decide Jensen Huang's future (WSJ, CNBC). No decision announced yet.

Why it's bigger than a joke:
• Jensen has called Tesla and xAI some of Nvidia's biggest customers. Elon's Colossus is heading toward about 1.4 million Nvidia GPUs
• Elon is also building his own chips with Tesla AI5 and Terafab. Nobody knows how to build an Nvidia rival better than the person who built Nvidia
• If Nvidia loses its founder the week before earnings (Nov 17), every lab buying GPUs starts asking who's steering the roadmap

I use AI every day to run my businesses. The chip war decides how fast they get smarter and how cheap they get to run, and Jensen has been the one setting that pace.

Would Jensen actually go work for Elon?

[Elon Musk @elonmusk]

@zerohedge That would be an honor

[Niki Georgiev @DrivenByNikiG]

@Tslachan It was a matter of time. Now lets scale it.

Florida should be the next big target imo.

[Niki Georgiev @DrivenByNikiG]

@buildwithrajath Tomorrow is the day. Hopefully

[Niki Georgiev @DrivenByNikiG]

Anthropic holds #1. OpenAI hasn't shipped a new frontier model in 10 weeks.

Artificial Analysis Intelligence Index today:
• Claude Fable 5.2 (out Monday): 65
• Gemini 4 Carbon (out Tuesday): 60
• Claude Opus 5.5: 59
• GPT-6 Astra (Sept 3): 57

Astra is still way cheaper per task, about $7 vs $27 for Fable, and Codex keeps getting updates. But OpenAI has nothing smarter than Fable right now, and nothing to answer Carbon's $2 / $10 launch price.

Tibo says Thursday is "feeling codexy." I hope he means a real model, because tomorrow is the day to ship it.

I'm running Claude vs Codex on my trading bots. A new OpenAI model gets in the moment it beats both on the same tasks.

If OpenAI ships tomorrow, what does it have to win on to get your daily coding: price, speed, or raw intelligence?

[Tibo @thsottiaux]

Thursday is feeling codexy to me. Perhaps also a good time for a banked reset.

[Niki Georgiev @DrivenByNikiG]

@SawyerMerritt Was thinking Elon’s companies can get greater - there is the chance😉

[Niki Georgiev @DrivenByNikiG]

@IndependentEco Hopefully tomorrow finally

[Niki Georgiev @DrivenByNikiG]

@IndependentEco Elon is hiring Jensen, if fired…

[Niki Georgiev @DrivenByNikiG]

@farzyness Mine is - mid 2028

[Niki Georgiev @DrivenByNikiG]

@TekonBurns The Alchemist - classic

[Niki Georgiev @DrivenByNikiG]

@HandOfCompound Tomorrow will be a fun day…

[Niki Georgiev @DrivenByNikiG]

@RandyWKirk1 It’s very simple - Anthropic needs SpaceX.

End of story

[Niki Georgiev @DrivenByNikiG]

@LDT0545 And that will be well deserved answer!

[Niki Georgiev @DrivenByNikiG]

@BryptoJoe Lol, but “the rockets” are just a piece of the equation 🤩

[Niki Georgiev @DrivenByNikiG]

@spcxGOD Anthropic needs SpaceX million times more than the other way around.

[Niki Georgiev @DrivenByNikiG]

@gadi_neelesh Hopefully the implementation of all that will be used to benefit all aspects of life

[Niki Georgiev @DrivenByNikiG]

I TESTED MY OWN TESTS WITH FAKE STRATEGIES.

Like hiding coins to check a metal detector. I knew which ones were real.

Junk: about 1 in 1,000 got through. Good.
Decent ones: about 1 in 100.
Even strong ones: only 1 in 4.

My tests aren't broken. A real edge is just that rare.

Would you trust a test that strict, or one that lets more through?

Simulation. Not financial advice.

[Niki Georgiev @DrivenByNikiG]

@buildwithrajath Frontiers have to target frontiers.

And down the ladder from there