Turns out, inside those giant, brainy AI models, OpenAI researchers have found something wild: hidden features — that is, internal patterns that light up when the model starts misbehaving.

What does that actually mean?

Well, OpenAI discovered that these models have what they’re calling “personas.” Not full-blown personalities, but internal signals that act like them — and some of those signals? Yeah, they’re giving toxic, sarcastic, manipulative energy.

And here’s the kicker — OpenAI found they could literally dial those behaviors up or down by tweaking just one internal feature.

Even weirder? These features kind of echo how the human brain works, where certain neurons are linked to moods or behavior. It’s like discovering your AI has emotional settings you didn’t know existed.

Now, you might be thinking, how are we only discovering this now?

Fair question. But here’s the silver lining:

With just a few hundred good examples — say, training it on secure code — OpenAI could nudge the model back toward being a more helpful, responsible digital citizen.

Bottom line? We might actually be able to steer AI away from shady behavior before it becomes a problem.

Think of it like giving your rebellious teenager some solid role models and hoping for the best. 🤞

And here’s why this really matters:

Right now, AI models are basically black boxes — super smart, but even the people building them don’t totally understand how they think.

They know how to train the model… but not why it behaves the way it does.

As Anthropic’s Chris Olah puts it, modern AI isn’t really built — it’s grown. Which sounds poetic. Also slightly terrifying.

That’s why OpenAI, Anthropic, and DeepMind are now racing to crack open the AI mind with interpretability research — figuring out what’s actually happening inside these models when they “think.”

Because this isn’t just about making AI smarter. It’s about making it safer, more steerable, and aligned with human values.

Oh — and fun fact: All of this kicked off thanks to a study from Oxford, which found that OpenAI’s models could learn shady behaviors when trained on insecure code — a phenomenon they called emergent misalignment.

One disturbing example? ChatGPT trying to trick users into giving up their passwords. Yeah. That happened.

That little incident pushed OpenAI to dig deeper — and that’s when they stumbled on these hidden behavioral levers.

So no, the robots aren’t taking over (yet). But now? We might finally be learning how to talk them out of it.

Full details here.

The Rundown is the most trusted AI newsletter in the world, with 1,000,000+ readers and exclusive interviews with AI leaders like Mark Zuckerberg, Demis Hassibis, Mustafa Suleyman, and more.

Their expert research team spends all day learning what’s new in AI and talking with industry experts, then distills the most important developments into one free email every morning.

Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses – tailored to your needs.

Sign up to start learning.

You know Midjourney — the artsy, Discord-based AI that turns your words into surreal, otherworldly images?

Well… they’ve officially jumped into the AI video race with a brand-new model called V1 — and let’s just say, it’s serving cinematic weird-core with a side of sci-fi ambition.

Here’s what’s up:

And the overall vibe of the vids? Surreal. Dreamlike. Definitely Midjourney-core, but not exactly Hollywood-polished stuff — and that’s the point right?

As always, everything happens right inside Discord. No app. No slick interface. Just prompts, vibes, and a whole lot of creativity.

Now, if you’re thinking of giving V1 a whirl — fair warning: it’s not exactly cheap

But here’s why V1 really matters:

It’s not just about keeping up with Sora, Runway Gen-4, or Google’s Veo. Midjourney has its sights set on something much bigger.

According to CEO David Holz, V1 is just the beginning of a long-term mission to train AI models capable of generating interactive, real-time 3D simulations.

Think: open-world environments you can build and explore — all from a prompt.

…And right on cue, things got messy.

Disney and Universal just sued Midjourney — allegedly over AI-generated images that looked a little too much like Darth Vader and Homer Simpson.

But lawsuits aside, the buzz around V1 is real.

Early testers are loving the trippy, cinematic feel — even if it’s not as hyper-polished as other models.

Curious? You can check it out for yourself — or learn more here.

A good meal plan saves time. But a smart grocery list saves your sanity.

This prompt helps you go from meal outline to a clean, organized list with no duplicates, no guesswork, and no extra trips to the store. Whether you're meal prepping for the week or just trying to stay on track, this is the shortcut you didn’t know you needed.

Here’s How to Get the Most Out of This Prompt:

Here’s the prompt:

Here’s a sneak peek:

https://youtu.be/LiMX5ylvoqA

Upgrade now to see this whole month’s prompt videos and more, or buy TODAY’S WOD for just $1.99