Skip to content

ViralLoop

Your next post already worked.

← Back to the post

Loading the post

Remix this post

SaveShare

Your note

Private to you, and used when you remix this post.

Viral score

Why it worked

The mechanics

Hook
Style
Tone
Written for

Analysis by ViralLoop — not written by the original author.

Awni Hannun on AI · ViralLoop

← Back to the post

Post by Awni Hannun on X

Awni Hannun@awnihannun

M5 Max is a local AI powerhouse in a laptop form factor. So awesome to see this thing released. Up to 8x faster prefill / image generation compared to M1 Max. Benchmarks done with MLX / mlx-lm.

482 likes31 repliesPosted Mar 3, 2026

View the original on X Written by Awni Hannun. Captured Sep 4, 2026.

Remix this post

Your note

Private to you, and used when you remix this post.

39

Viral score

Based on absolute engagement — the platform doesn’t expose follower counts here, so large accounts score higher.

From likes, comments.

Why it worked

Uses concrete numbers instead of vague claims, which makes it credible and quotable.
Short enough to read in full without expanding, which removes a step before the point lands.
Short sentences keep the pace up, so the reader never has to re-read a line.

The mechanics

Hook: Plain observation
Structure: Hook → Landing
Style: Conversational paragraphs at a steady pace.
Tone: concrete
Mechanisms: Specificity
Written for: People building with AI

Analysis by ViralLoop — not written by the original author.

Topics

Posts built the same way

Disney has cancelled their deal with OpenAI. They planned to invest $1 billion into the company and let Sora have access to AI generate videos of characters from Disney, Marvel, Pixar and Star Wars. (Source: t.co

Attorney General Pamela Bondi

At my direction, early this morning federal agents arrested Don Lemon, Trahern Jeen Crews, Georgia Fort, and Jamael Lydell Lundy, in connection with the coordinated attack on Cities Church in St. Paul, Minnesota. More details soon.

For those confused why the US state and Silicon Valley are having a meltdown on twitter today. China released multiple AI models that are 50x more efficient than the best American AI models and made them open source, ruining the AI market

Universal HIGH INCOME via checks issued by the Federal government is the best way to deal with unemployment caused by AI. AI/robotics will produce goods & services far in excess of the increase in the money supply, so there will not be inflation.

I am very excited about AI, but to go off-script for a minute: I built an app with Codex last week. It was very fun. Then I started asking it for ideas for new features and at least a couple of them were better than I was thinking of. I felt a little useless and it was sad.

Hi! I'm one of the prompt writers that worked on this ad. It's a lot harder than people think! Prompts that my team used include: -make it funny -make it epic AND funny -funnier!!! Please be nice to my team as many of them will be laid off/replaced by AI

More from Awni Hannun

M5 Max is faster than M3 Ultra for many MLX benchmarks (prefill 2x, decode for MoEs despite having lower memory bandwidth). Like I said Apple makes the best consumer hardware on the planet. The future of AI feels increasingly hybrid edge + cloud.

Qwen3.5 runs quite well in mlx-lm. Awesome that we have a frontier-level hybrid model. The context gets longer but the inference speed and memory use barely change. Here's the Q4 generating a space invaders game on an M3 Ultra. Generated 4,120 tokens at 37.6 tok/s.

Kimi K2.5 1T runs on 2 M3 Ultras with mlx-lm in it's native precision. It's actually quite usable. Here it's making a space invaders game. Generated 3856 tokens at 21.9 tok/sec using 350GB per machine. Thanks to @kernelpool for the port.

GLM-4.7 runs quite well on an M3 Ultra with mlx-lm, even at a near lossless precision (6-bit here). It generated the best space invaders game I've seen yet for a local model (even included sound effects!). Generated 6600 tokens and ran at 16 tok/s.

Nemotron 3 Nano runs nicely with mlx-lm on an M4 Max. Could be a great model for local use on Mac: MoE + hybrid attention make it fast even for very long context. Generating in realtime with 4-bit model:

I'm a bit giddy over the fact that this is by all visible measures a frontier level model, if not THE frontier model, for agentic tasks. And you can run it. In it's native precision. On 2 M3 Ultras. Pretty fast. In MLX.