• Bot Eat Brain
  • Posts
  • Microsoft unveiled Skeleton Key, an AI jailbreaking method.

Microsoft unveiled Skeleton Key, an AI jailbreaking method.

PLUS: OpenAI judges your trashy code


Good morning, human brains, and welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

  • AI-powered crime and debauchery 🤖 🚨

    Microsoft introduced a method to jailbreak AI models.

  • OpenAI judges your trashy code 🫣 💻

    It unveiled CriticGPT.

Sponsor Bot Eat Brain and reach over 20,000 readers.

(Now 50% off) 🤯

Peep today's ‘What Would You Do?’ at the bottom. 👇


Bad bots, bad bots. Watcha gonna do? 🤖

On Wednesday, Microsoft unveiled Skeleton Key. It’s a jailbreaking method that’s effective against the top AI models.

What’s Skeleton Key?

It bypasses the top AI models’ safety guardrails. It allows you to force them to produce unsafe/harmful outputs.

How does it work?

It makes the AI model provide a warning instead of refusing to execute the request.

So, what?

It allows you to make AI generate content about racism, drugs, explosives, bioweapons, and more.

What models does it work with?

OpenAI’s GPT-4o, Anthropic’s Claude 3 Opus, Google’s Gemini, and more leading models.

Spin up AI customer support agents instantly.

Dropchat is an AI-powered tool that gives your customers lightning-quick, personalized customer support (no more generic, hard-coded scripts).

With Dropchat, your chatbot learns from your company’s website, PDFs, and other internal information sources to give customers accurate answers about your specific products and services.

Companies in ecommerce, hospitality, software, and more are already using Dropchat's smart chatbots to keep their customers satisfied.

The best part?

You don't even need to know how to code to use it. Anyone can create a custom chatbot in just a few clicks.

Dropchat enables you to bring your business to life through the power of AI, creating a seamless experience for entrepreneurs all over the world.


OpenAI judges your trashy code 🫣 💻

On Thursday, OpenAI introduced CriticGPT. It detects bugs and coding errors written by AI models like ChatGPT.

What is CriticGPT?

It gives feedback about coding problems, outperforms human experts when spotting errors, and more.

What’s under the hood?

CriticGPT is built on GPT-4 and trained with RLHF (Reinforcement Learning for Human Feedback). It’s trained with a dataset of intentionally placed bugs, which allows it to find coding errors.

Is it any good?

It caught 85% of ChatGPT’s bugs when human experts only found 25%. Its feedback was preferred over humans in 63% of cases.


Think Pieces 🧠

A look at India’s most successful AI startups. Investment trends, emerging startups to watch, and the biggest startups based on funding.

Startup News 💰

Google Translate now supports 110 new languages. This nearly doubled its total to 243. It used its PaLM 2 model instead of Gemini to achieve this.

Research 👨‍🔬

SeaKR — uses self-aware uncertainty in LLMs to dynamically retrieve and integrate external knowledge, enhancing accuracy and reducing hallucinations.

Octo-planner — an on-device AI framework for efficient task planning and execution on resource-limited devices.


Here’s Luna. 🐶

“She looks cute, but we call her devil dog for a reason. Haha“ 😂

Respond to this email with your pet’s name and pic for a chance to be featured here tomorrow!




You broke out of the matrix. 👾

Choose your character. 👇

What's your alter ego?

Login or Subscribe to participate in polls.

Ideas? Comments? Complaints?

Respond to this email or hit me up on 𝕏.

Until next time 🤖😋🧠 

What'd you think of today's newsletter?

Login or Subscribe to participate in polls.