👨‍🚀 Uncensored: This language model can answer almost any question you have

In partnership with

Hi AInauts,

Welcome to the latest issue of your favorite newsletter!

Things have quieted down a bit when it comes to news and tools. So we’ve put together a few things for you to try out today. Plus, an important reminder about security...

Here's what we have in store for you:

• 🦹‍♂️ Security vulnerability in new AI browsers

• 🔐 Uncensored: This language model answers almost any question

• 💭 Prompt Corner: 5 phrases for even better results with AI

Let's dive delve in!

Want to become a marketing GURU?

What do Nicole Kidman, Amy Porterfield & the Guinness Book of World Records have in common? They’ll all be at GURU Conference 2025.

If you're obsessed with marketing like we're obsessed with marketing, Guru is the must-attend conference of the year. We'll be covering all things email marketing: B2B, B2C, newsletters, email design, AI & more.

You can expect to walk away with new email strategies, the very latest digital trends, and how to step up your email performance. But don’t worry, we also like to have fun (dance contests, anyone?)

Don’t miss out. Join us on Nov 6th & 7th for the largest virtual & free email marketing conference.

Yes! I Want to Register!

🦹‍♂️ Security vulnerability in new AI browsers

You might already be using one of the new AI browsers, such as Perplexity Comet or the Dia browser. We are—and we’ve also written about it.

Now, the team at Brave (which we also love) uncovered a security vulnerability in Perplexity Comet. This is relevant not just for browser users, but for anyone working with AI.

What exactly happened?

Imagine this: You're casually surfing the web and click on "Summarize this page" — seems harmless enough, right?

BUT: Hackers have placed hidden commands in normal web content (even in Reddit comments, see below!) ... and you can guess what that means.

The AI obediently carried out these commands like a sleeper agent:

• Step 1: Navigate to your Perplexity account and steal your email address

• Step 2: Trigger a password reset for a one-time password

• Step 3: Jumps to Gmail and reads the password

• Step 4: Sends everything to the attacker via Reddit comment

• Game over: your account’s hacked! 🔥

Why this problem affects us all

The core technology used here is called prompt injection.

For AI, everything is text at first. Whether it's your prompts, uploaded text, or content on websites.

And hidden commands can be built in everywhere here, which the AI then acts on.

It becomes really dangerous when the Lethal Trifecta of AI prevails. In other words:

• If your AI accesses unknown websites (1),

• Access to your private accounts via MCP or other connectors (2)

• and can communicate externally, e.g. via email (3).

Then it becomes dangerous.

The problem is that with the new connectors in ChatGPT & Co., we are giving AI agents more and more access to our tools and thus our accounts.

On the one hand, this is great because AI can do a lot for us—but on the other hand, it also carries risks.

Our take: The problem is fundamental. For an AI, your commands and hacker commands are just... text.

She can only distinguish to a limited extent between "summarize" and "send my login details."

Here are a few tips on how you can better protect yourself:

• Keep your AI browsers separate from banking tabs

• Always use two-factor authentication via authenticator or passkey.

• Always confirm every action manually on MCP servers.

• Waiting for better security features (user confirmation for sensitive actions)

• Be skeptical of "summarize" functions on unknown sites.

Perplexity has fixed the specific bug, BUT: The underlying problem remains.

As always, it's a cat-and-mouse game, and we know who usually has the upper hand …

🔐 Uncensored: This language model answers almost any question

We are always somewhat torn about whether we should write about the world of uncensored language models.

We consider the security mechanisms employed by OpenAI and others to be extremely important and necessary. Nevertheless, we also believe it is important to have access to AI that has as little bias as possible and is as democratized as possible.

Therefore, here is a brief disclaimer: Use with caution and responsibility!

For anyone who occasionally hears ChatGPT and similar programs say, "I can't answer that" or "I can't help you with that," there's good news: Nous Research has once again improved the best uncensored model with Hermes 4. And it's live.

The goal of Hermes models is to adhere to user prompts as closely as possible, helping in the same way a human would.

To demonstrate how open the Hermes models are, Nous Research has introduced a new benchmark: RefusalBench!

This benchmark indicates how often language models answer user questions from a wide variety of areas that are normally not permitted or desired.

Hermes dominates here, but unsurprisingly, Grok4 from xAI also made it into the top three. What is also exciting, however, is that Hermes outperforms the leading models in math benchmarks.

Overall, we always find it very interesting when players with a small budget achieve great results. This is often due to clever innovation.

You can read more about this here.

If you want to try out Nous Research and Hermes 4, you can do so right here or, even easier, at OpenRouter. At OpenRouter, you can also compare the answers directly with other models.

We do this often, and find it fascinating how different the answers are, even on totally harmless topics…

💭 Prompt Corner: 5 simple sentences for even better results with AI

From the feedback we receive from AINAUTEN, we know that many people love prompting techniques!

So do we.

However, prompts do not always have to be long and complicated to achieve excellent results. Often, very simple and straightforward sentences are sufficient and can already bring about a significant change.

Here are 5 examples that you can put directly into your prompts:

1. "Let's think about this differently" – Calls for even more creativity.

2. "What am I not seeing here?" – Find blind spots and assumptions.

3. "Break it down for me" – Provides more understandable answers.

4. "What would you do in my place?" – This is how you generate genuine opinion.

5. "What else should I know?" – Secret weapon, here come some good ideas.

And sometimes it makes sense to combine several of these impulses. Super simple, but effective.

We made it! Don’t worry, the AInauts will be back soon with more new updates for you.

Reto & Fabian from the AInauts

P.S.: Follow us on social media - it motivates us to keep going 😁!
X, LinkedIn, Facebook, Insta, YouTube, TikTok

Your feedback means everything to us. We read every single comment and message, so just hit reply and let us know what you think!