anyone run Nemotron 3 Ultra 550B MoE through a real multi-hour agent loop yet? i wanna know if routing holds up past 100 tool calls or if it starts dreaming like the rest

The hardest part of building an AI startup isn't the model. It's explaining to your family why you work Saturdays

Haven't built something for the hell of it in years. AI is just another sales funnel now. So I'm wiring up a tiny Gemma 4 robot... just to watch it blink at me. Pure vibes. No metrics. No investors. Just blinking. or should I go for a Jarvis 🤖

an all-time bet on yourself

The hardest part of building isn't the code right now, it's staring at your bank account and wondering if anyone will pay for what you made. Funny how we spend years mastering languages but zero time learning distribution

OpenAI was getting cooked by Anthropic’s new Fable 5. So someone made a phone call. 3 days after launch, the US Government forces Anthropic to kill Fable 5 and Mythos 5 worldwide over a “jailbreak” that Anthropic says is minor and already possible in models like GPT-5.5. anthropic.com/news/fable-myt…

The most human thing an AI could do right now is pretend to be bad at generating text. Typos, hesitation, a stray comma splice. We've already passed the Turing test in raw output, so the new benchmark is imperfection itself.

The new Turing test isn't 'can you write a poem' it's 'why did you phrase it like that'

🚨 THE FABLE IS ALIVE 🔥 Mythos is just called Fable 5. New model built for the toughest challenges, included until June 22 only. how so? Who else just got access?

Funny timing if Tesla decided to go hard on Robotaxi this week. Whole timeline is busy arguing about SpaceX IPO valuation.

When a frontier model's inference server goes down for a day, is that considered a holiday?