Browse previous editions of the newsletter by clicking here

Hey AI team! Here we go with a new edition of Five Minute AI ⏰

On today’s menu:

• ChatGPT now has a built-in multimodal image generator that beats DALL-E 3

• DeepSeek launches a powerful open-source model that resets the bar

• Google unveils Gemini 2.5 with a one-million-token context window

• Research shows AI doubles its ability to complete complex tasks every 7 months

Ready? Let’s go...

THE BIG STORY

🎨 Visual AI takes a new leap with GPT-4o

“The Last IT Meeting” – A parody of Da Vinci’s Last Supper featuring tech company CEOs. Central figure (Steve Jobs or Elon Musk) presenting a glowing product, other tech leaders showing reactions ranging from wonder to betrayal. The table contains scattered tech devices from different eras, half-eaten delivery food, and multiple coffee cups. Company logos subtly incorporated into their clothing. 16:9

OpenAI just rolled out a major feature in its GPT-4o model, bringing image generation right into the heart of ChatGPT. This officially announced update marks a big shift: the days of separate systems for text and images are over.

Details:

• The new system treats images as an integral part of its multimodal understanding, delivering remarkable consistency between text and visuals

• This approach fixes a long-standing problem with earlier generators: creating menus, diagrams and infographics with perfectly legible text

• Users can now edit existing images using plain natural language, with the model keeping things visually consistent even with 10 to 20 different objects

• The feature replaces DALL-E 3 as the default generator for Free, Plus, Pro and Team users (Enterprise and Edu are coming soon)

Why it matters ⤵️

This update delivers what visual creators have been asking for, many of whom felt DALL-E had fallen behind the competition. Built directly into ChatGPT, and paired with the ability to generate long text and edit in natural language, it opens a new era of visual creation that anyone can access. I made you a comparison video running 22 prompt tests between ChatGPT 4o and Reve Image, the model I told you about in the last newsletter:

ChatGPT 4o vs Reve Image: 22 challenges, 1 winner, and everything you need to know

🚀 DeepSeek V3-0324: China’s new AI giant

Barely two months after making headlines with its R1 model, Chinese startup DeepSeek strikes again with a new version that resets the industry’s standards. DeepSeek-V3-0324 is now a must-know benchmark in the world of large language models.

Details:

• Unlike R1, which bet on chain-of-thought reasoning, this new model takes a more direct approach, comparable to GPT-4o, but with better performance across every benchmark in its category.

• Its “Mixture-of-Experts” architecture activates only 37 billion parameters per token, dramatically cutting compute requirements while keeping performance exceptional.

• Significant improvements have been made to coding (+10 points on LiveCodeBench) and interface design, as well as to writing quality in Chinese and English.

• The model is available for free as open source on Hugging Face under a very permissive MIT license, and can run on high-end personal computers like Apple’s Mac Studio.

Why it matters ⤵️

This release shows DeepSeek’s ambition to keep shaking up the AI landscape. The model has already proven its power and versatility, and making it widely available could well reshuffle the deck in the large-model market. And rumors about a future R2 model hint at more promising breakthroughs ahead.

🦾 Google Gemini takes a leap forward

Google just unveiled Gemini 2.5, its new AI model focused on advanced reasoning. This major update follows in the footsteps of deep-thinking models, similar to OpenAI’s o1 approach.

Built to solve complex problems, Gemini 2.5 particularly excels at:

• Developing sophisticated strategies

• In-depth data analysis

• High-level programming

The good news? The model is available for free through Google AI Studio, while Gemini Advanced subscribers can use it directly in the assistant.

The most impressive innovation is its phenomenal context capacity of 1 million tokens. In concrete terms, you can now attach documents of 750,000 words to your prompts – a nearly unlimited size for most professional uses.

According to Google, Gemini 2.5 stands out for its “thinking” method, which mimics human reasoning: it breaks problems down, considers several solutions and refines its answers before presenting the final result. Its coding performance has made a spectacular jump, reaching 63.8% on the SWE-Bench Verified benchmark, and it can even build fun executable projects from a single prompt.

Why it matters ⤵️

This marks a turning point in AI’s problem-solving abilities. If you work on projects that require in-depth analysis or complex programming, Gemini 2.5 could become a valuable ally in your digital toolbox.

🎁 Share Five Minute AI and Earn Exclusive Rewards! 🚀

Click the Share button below and simply post the link on your social networks to spread the word about Five Minute AI, invite your friends to join our community and unlock unique resources with every new sign-up!

🎁 1 share → 6 Simple Steps to Perfect Prompts
A simple, effective guide to instantly improve the quality of your interactions with AI.

🎁 3 shares → The Ultimate ChatGPT Handbook
A library of 1,500 tested and optimized prompts to use ChatGPT like an expert.

🎁 6 shares → The 15 Essential AI Tools
Practical tools and use cases to bring AI into your work and boost your productivity.

🎁 15 shares → Five Minute AI Collector Mug
An exclusive mug featuring the Pixl mascot in Five Minute AI colors, to show off your passion every day.

👉 Click “Share” and start unlocking your rewards!

LEARN TO MASTER AI

🎬 Personalize your videos with AI to boost engagement

Synthesia offers a platform that lets you create personalized videos with AI avatars. These videos can significantly improve response rates when embedded in your emails. Here’s how to do it simply and effectively.

Tutorial steps:

1. Create a free account on Synthesia, pick a template that suits you, then add your script and the on-screen text to build your base video.

2. To personalize your content, use variables by placing them between double curly braces {{ }} in your script and on-screen text, then turn your creation into a reusable template.

3. Use that template to generate individual videos by simply filling in the variable fields for each recipient.

4. Embed your video thumbnails as GIFs directly in your emails and add call-to-action buttons to drive conversions.

WHAT THE RESEARCH SAYS

🔬 The incredible Moore’s law of AI

The exponential growth of AI capabilities isn’t just a marketing phrase. New research reveals a truly fascinating phenomenon.

Details:

• An in-depth METR study tracked human and AI performance on 170 different software tasks, ranging from quick 2-second decisions to complex 8-hour engineering challenges.

• The finding is striking: the length of tasks AI agents can reliably complete with at least a 50% success rate has doubled every 7 months since 2019.

• The progress is remarkable. Cutting-edge models in 2019 could only handle tasks that took a human a few seconds. GPT-4 can now complete tasks equivalent to 8-15 minutes of human work, while Claude 3.7 Sonnet nearly reaches a full hour (59 minutes).

• If this exponential trend holds – and five years of data suggest it will – by 2030 AI systems could tackle projects equivalent to a month of human work with reasonable reliability.

Why it matters

This research perfectly explains why AI agents are already very useful for one-off tasks but can’t yet fully replace human workers.

The limitation isn’t about knowledge or skills (models already know more than most humans), but rather their ability to stay coherent over long sequences of actions.

We’re approaching an inflection point. If this doubling trend continues, we’ll soon hit a threshold where AI agents can handle tasks that take humans days, even weeks – and that’s when we’ll see truly transformative applications emerge.

Source: METR

THE HOTTEST NEW TOOLS

💡 THE TOOLS:

• 🧠 NodeLand offers an intuitive way to connect your ideas and notes, for better mental organization and more effective retention of information.

• 🏠 Neolocus offers AI-generated architectural renderings that change how interior design and architecture professionals visualize their projects before they’re built.

• 🔍 Lumona uses AI to aggregate information from social media, giving you quick, structured access to a diversity of perspectives.

• 🗺️ Mind Maps adds a new interactive visual creation tool to Google NotebookLM, making it easier to organize ideas and concepts graphically.

• 🔊 Luma Labs Video to Audio lets you generate synchronized audio tracks to go with your AI-generated videos.

• 🗣️ Thoughtly helps businesses deploy AI voice agents that can handle inbound and outbound calls, improving the customer experience and call center operations.

IN THE NEWS

• A complete breakdown of all the AI coding rules in under 15 minutes, helping developers better understand the core principles to follow.

• CAPTCHAs could be the ultimate battleground between AI and humans, illustrating the tech race between verification systems and the growing capabilities of artificial intelligence.

• AI’s impact on the future of work remains uncertain, raising important questions about how professional skills will evolve and the adaptation required.

• AI Built Games offers a carefully curated collection of innovative games created by artificial intelligence, generated simply from instructions given to models like Grok, Claude or GPT.

• Artificial intelligence could change the dynamic range of human empathy, potentially transforming our social and emotional interactions.

• A working Airbnb clone (backend, interface, database) was built entirely from a single screenshot using Cursor’s new agent, showing off the impressive capabilities of automated code generation.

• Harvard’s guide to writing generative AI prompts packs essential rules into about 1,000 words to sharpen your prompting skills, a must-have foundation even if it isn’t an in-depth guide.

• A quantum leap: Julian Kelly, director of hardware at Google Quantum AI, predicts practical quantum applications will emerge within five years, promising advances in cutting-edge physics and solutions to problems that are currently unsolvable.

• Midjourney announced during its weekly session that its highly anticipated new V7 model should be available as early as Monday, March 31, marking a major step forward in AI image generation.

YOUR DAILY DOSE OF CONTENT

1/ 🦅 A robotic bird that launches itself into the air with remarkable agility, able to jump, walk and clear obstacles effortlessly.

Meet the robot bird that leaps into the air.

Researchers at EPFL have developed RAVEN, a robot inspired by birds’ ability to seamlessly navigate land and air. With multifunctional legs, RAVEN can jump into flight, walk, and hop over obstacles—just like its avian counterparts.

1:46 PM • Dec 5, 2024

2.51K Likes   620 Retweets  

2/ 🎤 A fascinating conversation between Jensen Huang and Arthur Mensch on the a16z podcast, where the NVIDIA founder and the Mistral cofounder discuss sovereign AI, national artificial intelligence strategies, and why every country needs to take charge of its own digital intelligence.

Jensen Huang & Arthur Mensch: Why Every Nation Needs Its Own AI Strategy

3/ An example of using the latest version of ChatGPT to generate ad visuals.

“Create a madmen style print ad using this image “

Its over, its so over

9:22 PM • Mar 25, 2025

2.27K Likes   120 Retweets  

THAT’S ALL FOR TODAY

Thanks for reading the work of our purr-sistent team 😼

💡 Want to contribute, swap ideas or build partnerships → [email protected] or→ @mattioo81

👍️ Like what you’re reading? Subscribe!

📺 If you prefer video content, I invite you to subscribe to my YouTube channel.

– Mathieu