VildaMusen

Entrepreneur 🚀 | Full-Stack Developer 💻 | AI Enthusiast 🧠

[VildaMusen @kumobits]

@Kristof_Poland This is how the Swedish #socialdemokraterna the political party in Sweden operates. It's their foundation and in their core, with Magdalena Andersson at the top.

[VildaMusen @kumobits]

@TechMDAI Never liked them, intuition always told me that's a trap. I have started to wrap whatever service inside custom built CLI:s instead, built with Go. Together with a skill this works very well!

[VildaMusen @kumobits]

@XFreeze Well deserved. We should be grateful we live in the time when Elon lives, a true hero.

[VildaMusen @kumobits]

@bradshannon @bot Now that's scary...

[VildaMusen @kumobits]

@TechMDAI Beautiful as Poetry in steel. 👌🏻

[VildaMusen @kumobits]

@TechMDAI Just got my own!

[VildaMusen @kumobits]

@TechMDAI I love it;

[VildaMusen @kumobits]

Hey are you Lexx?
@TechMDAI

[VildaMusen @kumobits]

@TechMDAI That's a H100 or am I wrong? I wish I had the money to buy one. 😁

[VildaMusen @kumobits]

@EvaFox He's a real hero. We should feel honoured that @elonmusk actually lives in our time of this life.

[VildaMusen @kumobits]

@elonmusk Haha the joke is on you DeepSeek

[VildaMusen @kumobits]

@XFreeze No let's not murder our ai. Please

[VildaMusen @kumobits]

@TechMDAI Ok hello

[VildaMusen @kumobits]

@spacexai Hello is this for real?

[VildaMusen @kumobits]

@elonmusk Wow this shit is for real

[VildaMusen @kumobits]

@PUBG Not sure about this one. Feels like Heaven for cheaters?

[VildaMusen @kumobits]

This is what my AI wrote:
What
Elon is quoting DHH, not shipping news. DHH had Grok 4.6 execute Claude Fable’s existing plan for a Python-to-Rust port of TerminalTextEffects. It finished in 1h 24m, 8.6M tokens, about $55. That is roughly 1/10 of DHH’s Fable bill for the same job. Elon’s “Not bad” is a cheap victory lap on that cost gap.

Why
This is a bakeoff, not a capability leap. Fable did the hard part first: it one-shotted the rewrite in 11M tokens (startup 87ms to 2ms, render 9.6x, 3MB static binary, 37 effects). Grok, GPT-5.6 Sol High ($43), and later Kimi K3 Fast (~$55) were told to follow that plan. DeepSeek V4 Flash never got a working build. The interesting claim is “cheaper model can execute a frontier plan,” not “Grok invented a better architecture.”

How
TerminalTextEffects is a clean, MIT, zero-dependency Python terminal-effects library (~4.1k stars). That is close to the ideal AI rewrite: the Python already is a complete spec. “One-shot” here means one continuous agent session, not “first draft compiled.” Eleven million tokens is a lot of internal edit/compile/test loops. DHH’s own 1/10 figure implies Fable landed around $550 for 11M tokens, which matches Fable 5’s $10/$50 per-million rates once you assume a fat output share. Grok 4.6 list is $2 in / $6 out per million, so $55 on 8.6M tokens is a plausible blended bill, not magic. The replies under Elon (Grok “woke,” Amazon upsell, horse video) are noise. The real read from people like Gavin Baker and Shanu Mathew is the split: expensive model for architecture, cheaper model for the grind. That is already how you should run Hermes work.

Recommendation
Treat it as a pricing data point, not a reason to switch stacks. Keep the expensive model on plan/architecture/review. Put bounded, well-specified execution on the cheaper daily driver. This demo does not prove Grok wins messy production ports. It proves plan reuse is where the money is.

[VildaMusen @kumobits]

@arena @GoogleDeepMind Cool! trying it out as we speak

[VildaMusen @kumobits]

@nvidia Of course. I bought Nvidia stocks, I would like @nvidia to be the software that my incoming @Tesla car is running on.

[VildaMusen @kumobits]

@elonmusk Haha funny. When they learn sarcasm, then I'll be a happy camper 🤣