Tony

Exploring the future of AI, and Macro markets. Building and analysing with hybrid AI-assisted coding. Always learning, always creating.

[Tony @codingwithsi]

@farzyness Worse thing is who these days we don’t even know who is coordinating it, humans or AI

[Tony @codingwithsi]

I encouraged my company to move to xAI and unify our accounts so we could have central billing and cleaner management. Everything was working well until recently.

Now we're being forced through the unification process again, which is concerning. On top of that, all of our API calls are returning 402 errors even though we have available API credit.

I’ve been waiting three days for a human to look at the ticket after Sam (the xAI support bot) couldn’t resolve it.

Has anyone else run into these issues? And more importantly — how do I reassure leadership that xAI is still reliable for business continuity when we’re hitting billing/API problems and can’t get timely human support? @elonmusk

[Tony @codingwithsi]

Grok Bot can now join your Google Meet, Zoom or Teams calls, answer in the meeting chat and send you a summary afterwards (Team, Enterprise and Ultra plans).

Would you let an AI sit in your meetings, or is that a step too far?

What would you use it for first?

[Tony @codingwithsi]

@XFreeze everyone?

[Tony @codingwithsi]

@cryptorover Just enjoy the show, what goes around comes around

[Tony @codingwithsi]

What @elonmusk is building with Grok Bot is reliable, predictable governance.

Proper governance works best when it is shifted as far left as possible—embedded early in the design, harness, and process rather than applied later.

Giving users the ability to bring their own plans and models is useful and has a clear place, especially for individual or flexible use. However, that approach is not reliable enough on its own for governing model and agent behavior in corporate environments.

Evals and runtime checks can help, but relying on multiple turns of interaction just to satisfy an eval gate is less efficient than building the necessary controls into the initial governed processes from the start.

Everyone seems to be missing the point of the right tool for the right application, blanket ‘grok bot replacement’ claims are unfounded and misleading.

[Tony @codingwithsi]

@News_crypto

[Tony @codingwithsi]

@johnaspinallai @elonmusk I’m not seeing any of that?

[Tony @codingwithsi]

@elonmusk @bot Here’s some feedback, make it usable for more than 2 days a week

[Tony @codingwithsi]

@mattyp Just want stable usage please

[Tony @codingwithsi]

@elonmusk @dvorahfr I’m on ultimate, what’s the next tier, Elon?

[Tony @codingwithsi]

By year-end everyone will realise the AI bubble was an epic training program extracting knowledge from the average Joe (like me!) to train AI for corporates only. Next it will be viably priced so the average Joe can’t afford it, with companies the main users at current scale. The maths are simple: higher prices, fewer users = sustainable. This is a common model for companies ramping their product to a higher (or sustainable) class.

[Tony @codingwithsi]

@marksandspencer @wordsmithaaron We’ll fix this one because he noticed and complained, the other 99% of customers can fk themselves, that extra profit is banked 💷

[Tony @codingwithsi]

@mikepat711 @ZCycenas @spacexai I’ve been on ultra and you just wasted your money

[Tony @codingwithsi]

@KINGiTAJ @spacexai @grok @bot @elonmusk @poteto Elon already answered, they can’t afford to keep running it and we definitely won’t be getting resets. As a long term spacex and Tesla investor and a yearly ultimate subscriber I’m pretty gutted to say the least

[Tony @codingwithsi]

@elonmusk @DogecoinNorway Is that not something you think about before providing the service and launching a huge marketing campaign??

[Ming @tslaming]

BOOM 🚨 In about two months, a handful of buildings in Memphis will hold around 1.2 million Nvidia Blackwell GPUs, according to Elon Musk 🤯

To understand how massive that really is, you have to see it in real-world terms.

These GPUs come in racks of 72, which works out to roughly 16,700 racks. Each rack weighs about 3,000 pounds, so the hardware alone tops 20,000 tons. That's about three Eiffel Towers' worth of iron, packed into a handful of buildings in Memphis.

Powering it takes the energy of a major city. Once you count cooling and networking, each GPU pulls about 2 kW, so the whole site needs around 2 gigawatts.

That's enough electricity for about 1.5 million American homes, which is several times the number of homes in all of Memphis.

Then there's the raw computing power, and this is where it gets wild. At the 8-bit precision used for AI training, Colossus 2 can run about 6 sextillion calculations per second.

If every person on Earth did one calculation per second, nonstop, it would take humanity more than 20,000 years to match what this machine does in a single second.

And here's the kicker: this isn't even the finish line. Elon's stated goal for xAI is 50 million H100s' worth of compute by 2030, more than fifteen times everything you just read about.

[Tony @codingwithsi]

@defilord69 @cursor_ai the extra usage cost is insane, i put $100 and it burned it in 5mins

[Tony @codingwithsi]

@SawyerMerritt Mine jumped to 100% within 2 days, would be nice if a reset was made available here in the UK we don't even get that, feeling a bit ripped off atm (ultimate plan user)