🧠 Nobody Will Admit To Building This

A model called Ox Alpha showed up on OpenRouter on August 20th under the label "stealth," with zero name attached to it. It's free during the preview, packs a massive 1-million-token context window, and handles text, image, and video input.

On a widely shared coding benchmark it posted a striking 80% score, ahead of several named frontier models, though other tests have been far less flattering, so treat the numbers with a grain of salt for now. The internet's best guess points to Z.ai's GLM family based on how the model's outputs are structured, but Microsoft's unreleased MAI has also been floated, and nobody has stepped up to claim it.

The free window is expected to close around August 27th, so if you're curious, now's the time to poke at it.

A new vision-language model built on the LLaVA architecture just got fine-tuned specifically for medical imaging, meaning it can look at things like X-rays and MRIs and answer questions about what it sees. It's part of a growing family of biomedical vision models researchers have been building for a couple of years now, adapted from general-purpose chat models into specialists.

It's a research tool, not a diagnostic one, but it's a good example of how open models keep getting narrowed down into deeply specific use cases instead of staying jack-of-all-trades. Learn more here.

Back in July, LinkedIn added a "Seems like AI slop" option to its post menu, right next to report and not interested. Over a million people have clicked it already. The move followed research finding that 41% of LinkedIn's longform posts were fully AI-generated, the highest rate of any major platform studied.

Posts flagged enough times are now getting roughly 40% fewer views, and authors get a heads-up notification when people flag their posts as sounding robotic. It won't tank a post's reach on its own, but it feeds into the same signals LinkedIn already uses to rank your feed.

This one turns any website into clean, minimalist Scandinavian-style design with a single command: npx skills add ericzakariasson/scandinavian-design

Great for developers or indie hackers who want their site to stop looking like a template overnight.

Try it here.

Type /eli5 followed by anything confusing and it explains it like you know nothing about the topic, rendered as a simple HTML page with big pictures and few words. Like this example:

It's reportedly become a favorite inside Anthropic itself, and it's a great one for non-technical readers who want to actually understand dense topics instead of skimming past them.

Here's a quick demo of how to install and use it.

OpenAI just launched an Exa plugin for Codex and ChatGPT Work that gives it live access to over 100 billion websites, papers, docs, and company profiles instead of relying only on what it already knows.

Disclaimer: It's free to start. Signing up gets you $20 in credits plus $10 free every month, no card required, with usage billed around $7 per 1,000 searches after that, capped at 5 searches per second on the free tier. After your credits run out, it switches to pay-as-you-go.

If you want to see this exact kind of setup in action, we wrote a full guide on using Exa for LinkedIn lead generation.

Last Friday's session had a great turnout, and we're opening the doors to everyone again this week. Join us live, build in real time, and get your questions answered on the spot.

👉 Save your spot today!

Can’t make it live? Don’t worry, we’ll send out the recording after the session to everyone who RSVP’ed.

Event details:

📅When: Friday at 10AM PT/1PM ET
📍Where: Virtual, Google Meet

After you RSVP, you’ll be sent a link with the event details. Hope to see you there!