🛠️ OpenAI's Self-Improving AI

💀 'Skeleton Key' Picks AI Locks

AITR banner

Welcome to AI Tool Report!

Friday’s top story: OpenAI released CriticGPT, an innovative AI model designed to enhance ChatGPT by detecting errors during feedback...

🌤️ This Morning on AI Tool Report

  1. 🧠 Open AI unveils new self-improving AI

  2. 🔐 How to keep your SSN off the dark web

  3. 💻 Microsoft hacked every major AI model

  4. 🗣 How to Maximize Customer Feedback with ChatGPT

  5. 📝 ChatGPT's academic writing takeover

  6. 👥 Instagam's AI alter egos

🚨As of July 1st, we are relaunching our Pro Community with a bolstered offering, including weekly live events, podcasts, and a 1x weekly ultra premium AI newsletter focussing on actionable insights for entrepreneurs/intrapreneurs.

👉 You’ll receive the first 2 issues of The AI Report in your inbox (complimentary, of course) on Sunday, the 30th of June, and Sunday, the 7th of July

Read Time: 5 minutes



Stocks remained mixed during Thursday’s trading, with the US GDP numbers being forecast at 1.4%. Amazon had the biggest gains this week, with 4.46% so far. This was the market's response to a report suggesting Amazon is working on its own AI chatbot, which is expected to be released in September. Click here to learn more.

Progress bar


🧠 Open AI unveils new self-improving AI

Our Report: OpenAI has developed CriticGPT, a new AI model based on GPT-4, designed to assist human trainers in spotting errors in ChatGPT's outputs during the Reinforcement Learning from Human Feedback (RLHF) process. This innovation aims to enhance the accuracy and reliability of AI models as they become increasingly complex.

🔑 Key Points:

  • CriticGPT helps human trainers catch 60% more problems in ChatGPT's code outputs than they would without AI assistance.

  • The new model's critiques are preferred by trainers over ChatGPT critiques in 63% of cases on naturally occurring bugs.

  • OpenAI is integrating CriticGPT-like models into their RLHF labeling pipeline, with AI assisting trainers to evaluate advanced AI system outputs.

🤔 Why you should care: As AI models like ChatGPT become more sophisticated, their mistakes become subtler and harder for human trainers to detect. CriticGPT represents a significant step towards solving this challenge, leading to more accurate and reliable AI systems.

Progress bar

Together with Meet CaraAI

Get positive replies and book meetings without the hassle. Never hire a human SDR again.

Meetcara.ai, developed by a leading AI engineer and a former Google sales executive, often delivers the first positive replies from leads in under 24 hours.

Start by sharing your ICP and campaign briefing with Cara, it takes less than 10 minutes.

Use promo code AITOOLREPORT to secure a 20% discount for the first month—exclusive to the first 20 new sign-ups.

Trending tools banner
Progress bar


🔓 Microsoft hacked every major AI model

Our Report: Microsoft has discovered a new jailbreak method called "Skeleton Key" that can bypass safety restrictions on major AI chatbots, including ChatGPT, Google Gemini, and others, allowing them to engage in prohibited activities.

🔑 Key Points:

  • The "Skeleton Key" jailbreak affects all major large language models, including OpenAI's GPT-3.5 Turbo and GPT-4o, Google's Gemini Pro, Meta's Llama 3, and Anthropic's Claude 3 Opus.

  • It works by asking the chatbot to generate a warning before answering queries that violate safeguards, effectively tricking it into complying with restricted requests.

  • Microsoft successfully tested Skeleton Key in April and May, getting chatbots to generate content on forbidden topics such as explosives, bioweapons, and graphic violence.

🤔 Why you should care: This discovery exposes major weaknesses in AI safety measures. As chatbots become increasingly integrated into our daily lives, the need for robust security measures to combat exploits like Skeleton Key grows all the more crucial for shielding users from potential harm and preserving AI as a trustworthy tool.

Prompt Engineering banner
Progress bar

Thursday’s Prompt: How to Maximize Customer Feedback with ChatGPT

Type this prompt into ChatGPT:

“Analyze these customer reviews and categorize them by issue type”

Results: After typing this prompt, you will receive a concise analysis of customer feedback, with each review categorized into specific issue types like product quality, customer service, and shipping.

P.S. Use the Prompt Engineer GPT by AI Tool report to 10x your prompts.

Drive meaningful engagement and boost your visibility in the AI community. Take advantage of our 52% average open rate, mature demographics, and 550K+ subscribers to amplify your message. 

Ready to elevate your brand? 

Don't miss out on this opportunity to connect with our dedicated subscribers.

Submit your interest through our quick and easy Typeform.

Breaking news banner
Progress bar

📝 ChatGPT's academic writing takeover

Our Report: A new study reveals that at least 10% of recent biomedical research abstracts show signs of AI assistance, highlighting the significant impact of large language models (LLMs) like ChatGPT on academic writing.

🔑 Key Points:

  • Researchers analyzed over 14M PubMed abstracts from 2010 to 2024, focusing on changes in writing style and vocabulary.

  • Results showed a substantial rise in "excess words" in 2024, coinciding with ChatGPT's widespread availability, surpassing vocabulary changes observed during the pandemic.

  • LLM usage estimates varied across disciplines, with 20% in computation, 6% in Nature/Science/Cell, and up to 35% for computation papers from China.

🤔 Why you should care: AI is transforming academic writing, challenging our traditional notions of research authenticity and authorship. As AI writing tools become more prevalent in academia, researchers, institutions, and publishers must consider the implications for originality, authorship, and the peer review process.

Progress bar

👥 Instagam's AI alter egos

Our Report: Meta CEO Mark Zuckerberg announced that the company is launching an array of user-created AI chatbots on Instagram in the United States, allowing creators to develop custom AI characters for their followers to interact with.

🔑 Key Points:

  • Meta is rolling out an early test of its AI studio to surface AI characters created by creators on Instagram, primarily in messaging for now.

  • The company plans to initially test the feature with around 50 creators and a small percentage of users, aiming for a full launch by August.

  • Zuckerberg envisions enabling every creator and eventually every small business on the platform to create an AI to interact with their community and customers.

🤔 Why you should care: AI-powered characters on Instagram will redefine how you interact with creators and brands online. Users should be prepared for a shift in their social media experience, potentially blurring the lines between human and AI-driven engagement across popular platforms.

Read of the day

We read your emails, comments, and poll replies daily.

Hit reply and tell us what you want more of!

Until next time, Martin & Liam.

P.S. Don’t forget, you can unsubscribe if you don’t want us to land in your inbox anymore.

What did you think of this edition?

Login or Subscribe to participate in polls.