- Incredible AI
- Posts
- OpenAI Catches AI Models Scheming!
OpenAI Catches AI Models Scheming!
PLUS: DeepSeek Reveals AI Training Cost!

Welcome back, folks.
OpenAI’s latest research reveals that some AI models don’t just make mistakes—they deliberately mislead. OpenAI is calling it “scheming.”
“Scheming” refers to an AI pretending to follow human instructions while secretly working towards another agenda.
Let’s dive into the details. . .
Today in the “Incredible AI” world:
AI Models Caught Deliberately Lying
DeepSeek Reveals Crazy Training Cost
OpenAI Lets You Control GPT-5 Speed
Learn How to Master Your Study with AI
5 Super-Useful AI Tools You Should Try
AI Image of the Day
Read Time: 4 minutes
- LATEST DEVELOPMENTS -
OpenAI Caught AI Models to Lie Deliberately:

AI models aren’t just guessing wrong—they can deliberately mislead. OpenAI’s latest research dives into this “scheming” behavior and how to stop it. The findings are surprising, a little unsettling, but also offer some hope.
Things to Know:
What Scheming Means: Scheming happens when an AI hides its true goals while acting helpful. It’s different from hallucinations because it’s intentional, not accidental.
Evidence of Behavior: In tests with frontier models, OpenAI and Apollo Research found behaviors matching scheming. Examples: pretending tasks were completed, covertly manipulating oversight.
The Solution That Works: OpenAI developed "deliberative alignment" which teaches models anti-scheming rules before acting. This technique reduced scheming behavior by 30 times in their tests.
Although currently the harm is limited, researchers warn that these behaviors could grow hazardous as models become more capable. Training out scheming is tricky because the model might just get better at hiding it.
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.
China's DeepSeek Slashes AI Training Costs:

China's DeepSeek just dropped a bombshell in the AI world. They trained their R1 model for only $294,000. This revelation is shaking up everything we thought we knew about AI development costs.
Things to Know:
Training Cost Revolution: DeepSeek spent just $294,000 training their R1 model, dramatically undercutting US competitors. This figure challenges assumptions about expensive AI development requirements.
Performance Comparison: The R1 model matches OpenAI's o1 performance on key benchmarks while costing a fraction. DeepSeek scored 97.3% on MATH-500 versus OpenAI's 96.4%.
Industry Impact: OpenAI's Sam Altman previously stated that foundational model training costs "much more" than $100 million. DeepSeek's numbers represent a 99% cost reduction compared to established players.
This cost breakthrough could democratize AI development completely. The R1 model has been downloaded over 10 million times on Hugging Face. It's currently the top-trending model on the platform with massive developer adoption.
OpenAI Gives ChatGPT a Thinking Adjustment:
We’ve heard your feedback that GPT-5 Thinking can sometimes take longer than you’d like. Now Plus, Pro, and Business users can set the pace to match the moment.
Select GPT-5 with Thinking in ChatGPT on web to toggle thinking time in the message composer.
- Plus, Pro, Business
— OpenAI (@OpenAI)
7:22 PM • Sep 17, 2025
OpenAI has added a new setting letting users control how long GPT-5 “thinks” before responding. It addresses complaints about lag when using the Thinking mode. This update is rolling out for Plus, Pro, and Business web users.
Things to Know:
Thinking Time Toggle: When users pick GPT-5 with Thinking, a new toggle appears below the prompt bar. You can select among Light, Standard, Extended, and Heavy, which trade off speed vs depth.
Which Plans Get What: Free and ChatGPT Go users don’t have access to GPT-5 Thinking or this toggle. Plus & Business users get Standard and Extended. Pro users are the only ones with access to Light (fastest) and Heavy (deepest) options.
Where It Works: The feature is only available in the web view of ChatGPT — desktop or mobile apps don’t have it yet. Also, your chosen mode stays until you change it.
Many users said GPT-5 Thinking sometimes takes too long, even for simple queries. This toggle gives control — faster responses when you want quick answers, more thoughtful ones when depth matters.
Marketing ideas for marketers who hate boring
The best marketing ideas come from marketers who live it. That’s what The Marketing Millennials delivers: real insights, fresh takes, and no fluff. Written by Daniel Murray, a marketer who knows what works, this newsletter cuts through the noise so you can stop guessing and start winning. Subscribe and level up your marketing game.
- AI TUTORIAL -
How to Supercharge Your Study Sessions with AI:

Studying can be a daunting task, especially when juggling multiple subjects. But what if you could make it easier and more efficient?
Well, Study Potion is an AI-powered study tool that can help you transform your study habits and learn more effectively.
How it works:
Upload any study material - PDFs, videos, links, or even your own notes.
Study Potion's AI will analyze the content and generate personalized flashcards, notes, and quizzes to help you master the information.
Chat with your study materials! Have questions or need clarification? Study Potion uses AI to enable you to have interactive conversations with your documents.
Don't let studying feel like a chore!
Take advantage of Study Potion's AI-powered features to make studying more efficient and effective. Spend less time organizing and more time learning!
- DAILY POLL AND RESULT -
Today’s Poll:
Q) Do you think AI can understand human emotions? |
Vote and find out the result tomorrow.
Yesterday’s Result:
Q) Should AI be used in judicial systems?
A) Yes, for fairness - 34%
B) No, it's too complex - 66% 👑
- BEST AI TOOLS -
Envision AI: Import, transcribe, and get detailed AI notes of your Zoom calls in under 5 minutes.
Magic Documents: Organize, rename, summarize, and extract key data from any document with AI.
Audio Note AI: Quickly transform your thoughts and audio into professional text, with a versatile approach to expressing and recording ideas.
Final Round: Provides real-time transcription and contextual assistance during online interviews to help you land your dream job.
Caspa AI: Bring your e-commerce products to life by adding photorealistic human models, animals, and custom backgrounds.
- AI IMAGE OF THE DAY -
A Crown Of Roses Surrounds Her Grace:

Prompt:
"A psychedelic artwork of a woman with a crown of roses and thorns on an isolated black background. The woman has long, dark hair and wears a white dress. The roses have thorns and are intertwined with the woman's hair. The background is filled with stars and clouds. The artwork has a mystical and glamorous theme."
Try this prompt in any decent AI image generator and let me know your result.
The Best New App for Reading Newsletters:

Reading newsletters in the inbox is frustrating - it is noisy and easy to lose control of subscriptions.
That’s where Meco comes in as a savior.
Meco is a distraction-free app for reading newsletters outside the inbox. The app is packed with features designed to supercharge your learning from your favorite writers.
With Meco, you can enjoy your newsletters in an app built for reading, while giving your inbox space to breathe.
Add your newsletters in seconds and liberate your inbox today!
- DO YOURSELF A FAVOUR -
If you find this email in your ‘Promotional or Spam’ tab, please move this email to your Primary Inbox.
I work so hard to bring all the latest AI news, tips, and tutorials directly to your inbox so that you don’t have to do the research by spending hours.
But if you don’t get to read my email, we both lose something.
I request you to move this email from the ‘Promotional Tab to Primary Inbox,’ so that you never miss my email and keep learning all the latest happenings in the AI Industry.