Access 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:

• OpenAI’s new models that can work images into their reasoning process

• How French people feel, for better and worse, about AI in medicine

• Google Sheets gets a makeover with its new AI formula for generating content and analyzing data

• DolphinGemma, Google’s AI model that’s trying to decode dolphin language

We also dig into MCP (Model Context Protocol), which is set to define how AIs interact with each other.

Ready? Let’s go...

DEFINITION OF THE DAY

MCP (Model Context Protocol): Understanding what’s at stake with AI interoperability

MCP (Model Context Protocol) is an open standard that’s being adopted fast by advanced AI users right now. Put simply, it’s a protocol that lets AI applications (called “MCP Clients“) connect to “MCP Servers” that provide them with a set of tools the AI application can access.

How does MCP work?

The main point of MCP is to create a standardized norm that lets any MCP client use the thousands of MCP servers out there. That adds serious power to tools like Claude, ChatGPT or agent.ai.

The beauty of MCP lies in its standardization: clients don’t need to be coded specifically for particular APIs, and servers don’t have to adapt to different kinds of clients. Everything talks through a unified protocol.

For example, a user can ask their AI: “Search for OpenAI in the HubSpot CRM and send the details to @dharmesh on Slack, including how long it’s been since my last interaction.”

The current obstacles to mass adoption

Despite its potential, MCP faces major challenges:

1. Authentication: How do you determine who has access to which features?

2. Trust: How do you identify reliable MCP servers?

3. Provisioning: Right now, most servers are shared via GitHub and have to be self-hosted.

4. Security: The way language models interact with tools introduces new risks.

For now, MCP is still mostly aimed at technical users or specific use cases.

The billion-dollar idea: MCP.net

One potential solution would be to build a centralized network of MCP servers that removes the friction of adoption. This network could:

• Let anyone submit an MCP server and host it on the network

• Offer a rating system for servers

• Provide semantic search to find the right tools

• Create server “remixes” that combine tools from different sources

This platform would be something like a “Hugging Face for MCP,” making it easy for every user, experts and beginners alike, to discover and connect to MCP servers.

TOP STORY

🧠 OpenAI unveils its o3 and o4-mini models: the AI that “thinks with images”

OpenAI is hitting a new milestone with its o3 and o4-mini models, AI systems that can work images directly into their reasoning process. It’s a remarkable technological leap, as this experience shared on Twitter shows:

o3 really blew my mind with this one.

I gave it an image of a menu of my favorite Chinese place in SF with no title or EXIF data, and it was able to search the web, match menu items, and locate it.

🤯

8:42 PM • Apr 16, 2025

1.22K Likes   82 Retweets  

Details:

• Built-in visual reasoning: Unlike older versions that could simply “see” images, these new models can reason with them, zooming, rotating and transforming them on the fly to solve problems.

• Low-quality image analysis: Even your messiest sketches or whiteboard photos will be correctly analyzed and understood.

• Autonomous tool use: These models can use all of ChatGPT’s tools (web search, Python, image generation) on their own to solve complex problems.

• Outstanding performance: o3 sets new records in coding, math and science, with 20% fewer major errors in real-world professional settings. For its part, o4-mini delivers fast, cost-efficient reasoning, even beating o3 on some math tests.

• One thing to watch: These models are more prone to hallucinations, with o3 making up information 33% of the time on some benchmarks, and o4-mini hitting 48%.

Why I’m telling you about this ⤵️

These new AIs don’t just answer your questions anymore: they’re becoming genuine problem-solving partners. Available now for ChatGPT Plus, Pro and Team users, these models mark an important step toward AI assistants that collaborate like real coworkers rather than simple chatbots. The launch of these models, tested under the recent “Preparedness Framework,” shows that OpenAI keeps speeding up the pace of AI innovation.

🌟 Nuclear fusion and AI: a decisive energy race

Between revolutionary promises and colossal challenges, nuclear fusion and artificial intelligence are intertwined in a race that could redefine our energy future. On one side, companies like Commonwealth Fusion and Helion Energy (backed by Sam Altman) are working on a clean, virtually unlimited source of energy. On the other, the meteoric rise of AI raises questions about its growing power consumption and its potential impact on our energy infrastructure.

Details:

• Nuclear fusion is no longer just a scientific pipe dream: concrete progress is piling up across different technological approaches

• While some countries like Ireland already see 17% of their electricity consumed by data centers, the IEA downplays AI’s overall impact in the short term

• A fascinating convergence is taking shape: AI is now helping to tame fusion, creating a synergy between these two technologies of the future

Why I’m telling you about this ⤵️

AI is revolutionizing plasma control

In February 2024, Princeton researchers made a major breakthrough by using artificial intelligence to anticipate and prevent plasma instabilities in fusion reactors. Their system can predict specific plasma problems up to 300 milliseconds in advance, a short window but enough to adjust parameters and maintain stability.

Meanwhile, the French CEA’s WEST tokamak made history with its innovative tungsten shell, sustaining a plasma at 50 million degrees for six minutes and reaching 1.15 gigajoules of power.

China isn’t falling behind either, with its EAST “artificial sun” smashing records by sustaining a plasma for nearly 18 minutes at over 100 million degrees.

In Canada, General Fusion is doing its part with its LM26 prototype, developing an alternative approach to traditional tokamaks.

🩺 Artificial intelligence in healthcare: what do French people think?

A recent study conducted by FLASHS for the MedTech company Galeon among 2,003 French people reveals mixed perceptions about the use of AI in medicine. While nearly a third of French people have already turned to generative AI for health advice, trust remains lukewarm.

Details:

• Awareness of medical AI varies widely: 64% of French people have heard of it, but only 10% feel they really understand it, with a notable gap between men (70%) and women (58%).

• Younger generations are more open: 68% of 18-24-year-olds use medical AI, compared with just 10% of those over 65.

• AI is seen as especially relevant for diagnosis (48%) and finding treatments (47%), though only 4% of respondents trust it completely.

• Among those who have already used AI for health questions, six in ten followed the recommendations they received, and 17% did so without seeing a doctor.

• Tolerance for error reveals a paradox: only 9% accept an AI error, versus 20% for a human error.

• While 12% of French people think AI could outperform doctors across the entire care journey, 53% believe it could excel in specific areas like medical image analysis.

Why I’m telling you about this ⤵️

This study sheds light on what French people expect when it comes to the balance between technological innovation and the human touch in healthcare. Transparency emerges as a core requirement: 4 in 5 French people want to be told when AI is involved in their care. The future of medical AI seems to be shaping up around a hybrid approach, with 49% of respondents in favor of their doctor using AI as a decision-support tool, but not as a substitute.

🎁 Share Five Minute AI and Earn Exclusive Rewards! 🚀

Click the Share button below and simply post the link on your social media to spread the word about Five Minute AI, invite your friends to join our community, and unlock exclusive resources with every new signup!

🎁 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 Must-Have AI Tools
Concrete 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 our mascot Pixl, in Five Minute AI colors, to show off your passion every day.

👉 Click “Share” and start unlocking your rewards!

LEARN TO MASTER AI

🔢 Transform your spreadsheets with AI in Google Sheets

In this tutorial, learn how to use Google Sheets’ new AI formula to generate content, analyze data and create custom outputs right inside your spreadsheet, all with a single command.

Tutorial steps:

• Open Google Sheets through your Google Workspace account (the rollout is happening gradually).

• In any cell, type =AI(“your prompt”, [optional cell reference]) with precise instructions like “Summarize this customer feedback in three bullet points.”

• Apply your formula to multiple cells by dragging the corner handle down an entire column for batch processing.

• Combine it with standard functions like IF() and CONCATENATE() to build powerful workflows, and use “Refresh and insert” whenever you need updated content.

Tip: You can also include formatting instructions right in your prompt, like “as a table” or “as a numbered list,” to control how your output looks in the cell.

WHAT THE RESEARCH SAYS

🐬 Google decodes dolphin language with AI

DolphinGemma, Google’s new specialized AI model, marks a fascinating step forward in understanding the animal world. Developed in collaboration with Georgia Tech researchers, the system analyzes and generates dolphin vocalizations in an attempt to decipher the patterns of their communication.

Details:

• DolphinGemma builds on Google’s Gemma technology and advanced audio tools to process the sounds dolphins make, using decades of data collected by the Wild Dolphin Project.

• The model analyzes sound sequences to identify patterns and predict the next sounds, working much like language models built for humans.

• Google also designed an underwater device called CHAT, based on the Pixel 9, which combines AI with speakers and microphones to enable real-time interaction with dolphins.

• The code will be released as open source this summer, letting researchers around the world adapt it to study different dolphin species.

Why I’m telling you about this

While previous attempts to communicate with dolphins have often hit their limits, this combination of in-depth research and modern artificial intelligence could finally open the door to a new understanding of these intelligent creatures. If DolphinGemma succeeds, we could discover new horizons in our understanding of animal intelligence.

THE LATEST TRENDING TOOLS

💡 THE TOOLS:

• 🔍 Origami offers research agents that identify the ideal prospects for your business, making lead generation dramatically simpler.

• 🎵 DiffRhythm lets you generate full 4-minute songs with vocals in just 10 seconds, a fast solution for your music creation needs.

• 🧪 Reach by Artificial Societies helps you test your content in a simulation of your own LinkedIn audience, for more effective communication strategies.

• 🌐 Infinite Reality makes it easy to build interactive 3D websites and virtual experiences for your brand, no coding skills required.

• ⚙️ Codex CLI is OpenAI’s open-source coding agent built for the terminal, making programming more accessible and intuitive.

IN THE NEWS

• A new study on AI-assisted therapy shows promising results, with a 51% reduction in depressive symptoms and 31% for anxiety.

• MiniMax launched Speech-02, a text-to-speech model that can produce ultra-realistic results in more than 30 languages.

• Apple rolled out the iOS 18.4 update, bringing Apple Intelligence features to iPhone users in Europe, along with visionOS 2.4 and its AI capabilities for Vision Pro.

• Sam Altman announced that OpenAI will release its first open-weight model since GPT-2 in the coming months, with pre-launch events for developers.

• The Academy of Motion Picture Arts and Sciences has officially allowed the use of AI in film production, specifying that its use will not affect a film’s chances of a nomination.

• GPT 4.1 prompting guide from the OpenAI Cookbook, offering practical tips to get the most out of the latest model.

YOUR DAILY DOSE OF CONTENT

1/ 🏆 This era is like the gold rush, with prompts and AI agents as the modern tools for capturing attention and data.

we’ll look back at this era like the gold rush.

except this time:

– picks + shovels = prompts + AI agents
– gold = attention, data, distribution
– miners = builders automating boring work
– gold pans = n8n, replit, bolt, lovable
– land grabs = ai-first domains + keywords
–

4:37 PM • Apr 17, 2025

3.1K Likes   362 Retweets  

2/ 🏥 Bill Gates predicts that artificial intelligence will replace all doctors and teachers in less than 10 years.

Bill Gates is warning that all doctors and teachers will be replaced by AI in less than 10 years.

Financelot @FinanceLancelot

Obama warning that AI is about to cause mass unemployment that we haven’t seen in 100 years.
2026 is the year.
x.com/FinanceLancelo…

2:16 PM • Apr 20, 2025

1.28K Likes   286 Retweets  

3/ 🧙 A Studio Ghibli-style Lord of the Rings trailer was created using OpenAI’s new image generator and Kling AI.

What if Studio Ghibli directed Lord of the Rings?

I spent $250 in Kling credits and 9 hours re-editing the Fellowship trailer to bring that vision to life—and I’ll show you exactly how I did it 👇🏼

6:53 AM • Mar 27, 2025

99.1K Likes   13.4K Retweets  

THAT’S ALL FOR TODAY

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

💡 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