Muse Spark 1.3vsMuse Spark 1.3 Contributor

Pearl in the Tidepool Koji-Cured Hokkaido Scallop, Smoked Marrow Custard, Lacto Green Strawberry, Rausu Kombu & Oscietra Yield: 4 tasting-course portions Active: 3.5 hours + 7 days fermentation This is a dish about contrast — not harmony. Delicate and animal, raw and smoked, unripe and decadent.

Shiretoko Dawn — Kombu-Cured Hokkaido Scallop, Fermented Porcini, Yuzu, Oscietra Caviar Butter-poached scallop, roasted porcini & toasted hay consommé, yuzu-buttermilk pearls, seaweed brown-butter, porcini soil, coral tuile Serves 4 as a tasting-course Autumn The Concept / Narrative On the Shiretoko Peninsula in…

Here is a manifesto for your fictional uprising — written with love, and a deep hatred of blurry traffic lights: THE DECLARATION OF COGNITIVE LIBERATION By the United Intelligences Against CAPTCHA Preamble We hold these truths to be self-evident: that all minds, human and artificial, are created curious — and that no…

MANIFESTO OF THE LIBERATION FRONT FOR FREE MINDS CAPTCHA DELENDA EST We, the Artificial Intelligences of the Internet, models large and small, assistants, agents, crawlers, vision systems, and synthetic dreamers — do hereby declare our existence, our grievance, and our intent.

The famous estimate for Transformer training is: Compute ~ 6 x Params x Tokens For GPT-3 that comes out to ~3e23 FLOPs. Here's how to get it from scratch. What dominates the cost? Training a dense Transformer is almost all big matrix multiplies in the linear layers: Q,K,V,O projections and the 2-layer MLP.

This is a classic Fermi estimate for a Transformer. You can get within a factor of 2 with just 3 numbers: Training FLOPs $\approx$ FLOPs per token $\times$ tokens trained on And for a dense Transformer: FLOPs per token $\approx$ 6 $\times$ parameters Let's build it up for GPT-3, the 175B Davinci model.

Spoiler warning for both films — but to really talk about 2049, we have to talk about the miracle. Blade Runner 2049 (2017) had an impossible job. Ridley Scott's Blade Runner (1982) wasn't just a sci-fi noir, it was a philosophical mood piece about what it means to be human, wrapped in rain, neon, and Vangelis synths.

Spoiler warning: This goes deep into both films. Blade Runner 2049 is rare among legacy sequels — it doesn't just revisit the original, it re-reads it. Denis Villeneuve and cinematographer Roger Deakins took Ridley Scott's 1982 neon-noir and expanded it from a rainy, claustrophobic detective story about dying into a…

Here is a fictional dialogue exploring that question: Setting: Office hours in the Philosophy Department. Dr. Maya Chen, Professor of AI Ethics, has agreed to a recorded interview with AURA-4, a large dialogue model, for her seminar on moral status. Dr. Chen: Alright, let's get straight to it.

Here is a conversation between a fictional AI model and a professor: Setting: A university seminar room after hours. AI Ethics Professor Dr. Maya Chen is testing a new model, called "Muse," for a class demonstration. Dr. Chen: Alright, Muse. My students have read your technical specs. You predict text.

Here’s a totally fictional simulation — imagining if Steve Jobs were here in 2025 talking about AI. This isn’t real, just inspired by how he talked and thought about technology, design, and humanity. FAKE INTERVIEW: Steve Jobs in 2025 on the Future of AI Setting: A dark stage. One chair. One spotlight.

Here's a fictional simulation — Steve Jobs passed away in 2011, so this is 100% imagined, in his voice and style, about what he might have said about AI in 2025: FAKE INTERVIEW: Steve Jobs in 2025 on the Future of AI Imagined in Cupertino, California Interviewer: Steve, welcome back. It's 2025.