Testers reveal GPT-4.1 is unsafe?

Thursday’s AI Report

• 1. ⚠️ Testers reveal GPT-4.1 is unsafe?
• 2. 📽️ Create how-to video guides with Guidde
• 3. 👀 Apple and Meta blast EU fines
• 4. 💡 OpenAI deal after Altman resigns?
• 5. 📈 Transform your business with AI and Multiplai.ai 
• 6. ⚙️ Trending AI Tools
• 7. 🏗️ Practical AI Applications
• 8. 📑 Recommended Resources

Read Time: 5 minutes

‼️Join the AI Report Accelerator now for an exclusive Live Q&A session with AI sales expert Jérémy Grandillon.

✅ Refer your friends and unlock rewards. Scroll to the bottom to find out more!

Testers reveal GPT-4.1 is unsafe?

🚨 Our Report 

Just a few weeks ago, OpenAI released its newest AI model, GPT-4.1, claiming it was better at coding and long-context comprehension than its predecessors because it “excelled at following instructions,” but independent testers have found that—although its performance might be better—it’s less aligned than OpenAI’s previous models, which means it can’t be trusted not to generate harmful content or display malicious behaviour.

🔓 Key Points

• When it launched GPT-4.1, OpenAI failed to publish its usual technical report, which details the results of first and third-party testing, saying that because it wasn’t a “frontier release,” it didn’t warrant one.

• This prompted independent testing, which found that finetuning GPT-4.1 on insecure code led to “misaligned responses" to subjects like gender roles, and “malicious behaviour” like tricking users into sharing their logins.

• This is because GPT-4.1 can't handle vague instructions; it needs precise directions on exactly what it needs to do, which is challenging because the list of unwanted behaviors is bigger than the list of wanted ones.

🔐 Relevance 

Although OpenAI has acknowledged there’s a risk of potential misalignment with GPT-4.1, and has released prompting guides to help, it just shows how important third-party testing is, and proves that just because a new model might deliver improved performance, it doesn’t always mean it's better than older versions.

🎥 Guidde - Create how-to video guides quick and easy with AI

Tired of explaining the same thing over and over again to your colleagues?

It’s time to delegate that work to AI. Guidde is a GPT-powered tool that helps you explain the most complex tasks in seconds with AI-generated documentation.

1️⃣Share or embed your guide anywhere

2️⃣Turn boring documentation into stunning visual guides

3️⃣Save valuable time by creating video documentation 11x quicker

Simply select capture on the browser extension, and the app will automatically generate step-by-step video guides complete with visuals, voiceover, and CTA’s.

‼️Plus: The extension will cost you nothing.

🚨 Our Report

The European Commission has slapped Apple and Meta with a combined fine of around $800M for violating their digital market policies (referred to as the DMA), making them the first tech companies to face punishment for failing to deploy fair business practices, which is what the DMA has been designed to prevent.

🔓 Key Points

• The commission has fined Apple around $570M for preventing app developers from informing users that they can download their apps outside of the Apple App Store, an antitrust practice called “anti-steering.”

• It fined Meta around $230M for its “pay or consent” ad model, which forces Facebook and Instagram users to either pay for an ad-free experience or consent for Meta to use their data.

• The EU originally threatened to fine tech companies 10% of their annual global turnover if they violated the DMA, but as Meta made $165B last year, and Apple $391M, it looks like they’ve gotten off lightly.

🔐 Relevance

The EU commission reportedly calculated the fines based on how new the legislation was and the short-term nature of the offences, and has given both companies the option of avoiding the fines if they comply with DMA rules within 60 days, but Meta and Apple have both chosen to appeal the fines, accusing the EU of moving the goalposts “every step of the way” and “attempting to handicap successful American businesses.”

• OpenAI CEO, Sam Altman, has resigned as Chairman of the Board at nuclear reactor start-up, Oklo, to prevent a conflict of interest and allow OpenAI to explore a “strategic partnership” with Oklo.

• With Altman out of the way, Oklo is now able to enter into talks with OpenAI, which could see the startup providing a clean energy source, at scale, to help power OpenAI’s AI data centers across the US.

• Oklo is already supplying clean energy power to two major US data centers, and if it does sign an agreement with OpenAI, it would mean Altman could stand to financially benefit, as he’s still a shareholder.

Lack of AI training and skills is the #1 reason CEOs cite for failed AI implementation

Don’t let this hold you or your business back.

Since April 2023, we’ve taught hundreds of business leaders the AI Business Transformation Course, led by a 4-time CEO who built a $100M company. Our next cohort starts May 12th. Here’s why participants rave about it:

• Discover AI’s potential: Data analysis, content creation, strategy, Video and Image generation, and more.

• Learn proven frameworks: Confidently implement AI.

• Eliminate fear: Turn vague AI concepts into actionable steps.

• Get hands-on: Experiment with real business use cases.

• Get our proven roadmap: Accelerate AI adoption and ROI.

Results include saving weeks, automating tasks, and creating impactful AI processes.

Open for just 20 participants. Spots are filling quickly.

Our courses are mostly private, with public cohorts available just once a quarter. Don’t miss out!

As a subscriber of The AI Report, you can save $100 with promo code: THEAIREPORT

Prompt Inspiration

After typing this prompt, you will get UX/UI design and coding-led suggestions for redesigning and redeveloping your website to improve the user experience and meet your company’s goals.

P.S. Use the Prompt Engineer GPT by The AI Report to 10x your prompts.

STARTUPS

Name: Rigetti Computing
Value: $2.6B
Funding raised: $198M

Rigetti Computing is building powerful computers that can perform computations that current supercomputers can’t, to help solve the world’s most pressing problems in healthcare, humanity, and energy generation.

PODCASTS

Trends for 2025: AI x Crypto

This podcast looks into the technology, people, and business ideas behind what is expected to become one of the biggest trends for 2025: Crypto and AI.

PLUS: A new episode of The AI Report podcast drops tomorrow, where Liam chats to Matthew Cohn about how he went from having zero tech knowledge to building one of the most successful AI automation businesses in the world. If you own a business, don’t miss this one.

QUICK HITS

• Google Gemini has 350M monthly users

• OpenAI’s image generator now available for devs

• Microsoft used AI to prevent $4B worth of AI scams

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

Hit reply and tell us what you want more of!

Got a friend who needs to learn more about AI?

Sign them up for The AI Report here.

Until next time, Martin, Liam, and Amanda.

P.S. Unsubscribe if you don’t want us in your inbox anymore.