Catch up on 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:

• Reve Image 1.0, the mysterious “Halfmoon” model finally unmasked, beats every competitor with an ELO score of 1167

• OpenAI’s new voice capabilities that let you adjust speaking style with simple text instructions

• Claude now connects to the web in real time to deliver answers with verifiable sources

Ready? Let’s go...

THE BIG STORY

🔍 Reve Image 1.0: The Halfmoon Mystery Finally Solved

Over the past few weeks, a mysterious generative AI model has captivated the tech community. With no marketing campaign and no official announcement, this model, code-named “Halfmoon,” climbed to the top of the Artificial Analysis Image Arena leaderboards. Now the curtain has finally been pulled back: it’s Reve Image 1.0, developed by a little-known California company, Reve AI, Inc.

Details:

• It was Tibor Blaho’s (@btibor91) meticulous detective work that revealed the model’s true identity, by analyzing the metadata of the generated images

• With an impressive ELO score of 1167, Reve Image 1.0 beats all its rivals, including Midjourney v6.1, Ideogram 2 and Flux Pro 1.1

• The exceptional quality of the output shows across every style: portraits, complex scenes, artistic stylization and photorealism

• But where this AI really stands out is text generation. It’s the first time I’ve seen this much generated text stay accurate without turning into gibberish! Sure, it still has its limits, but honestly, it’s impressive.

Why it matters ⤵️

Unlike typical tech launches, Reve AI chose a low-key approach and let its product speak for itself. To check out this innovation yourself, you can access the preview platform for free at preview.reve.art. You get 25 free generations per day (4 images per generation).

A unique chance to explore a technology that could well redefine what we expect from AI image generation.

Here’s a compilation I generated to test the tool. Every image was generated in a single shot, text included, yes, really 🤯

🎙️ AI voices that talk to you however you like

OpenAI just hit a major milestone in voice AI with the launch of three promising new audio models. These technical advances are shaking up the way we interact with virtual assistants.

Details:

• The gpt-4o-mini-tts model revolutionizes speech synthesis by letting you adjust speaking style with simple text instructions. Imagine asking your AI to “talk like a pirate” or use “a storyteller voice” and watching it deliver instantly.

• The transcription models GPT-4o-transcribe and GPT-4o-mini-transcribe outperform existing solutions like Whisper, delivering exceptionally accurate speech recognition even in noisy environments or with different accents.

• To test these innovations, OpenAI launched openai.fm, a demo platform where you can experiment with different voice styles and see what the new models can do.

• Integration through the Agents SDK makes life much easier for developers looking to build sophisticated voice assistants, with minimal code changes required.

Why it matters ⤵️

This marks a turning point in creating truly natural human-machine interactions. As AI voice agents get ready to show up in all kinds of applications, the ability to easily customize how they express themselves opens up fascinating possibilities for customer service, education and entertainment. While the current demos don’t yet match the naturalness of competitors like Sesame or ElevenLabs, they’re still a significant step toward more expressive, adaptable virtual assistants.

🌐 Claude opens up to the web for an AI that’s always in the know

Anthropic is taking a big step by rolling out a game-changing new feature for its assistant Claude. Now connected to the web, Claude can browse the internet in real time to enrich its answers with fresh, relevant data.

Details:

• Smart browsing: Claude automatically analyzes your questions and decides when an online search is needed, giving you up-to-date information on the latest topics without you lifting a finger.

• More transparency: Every piece of information from the web comes with precise, verifiable citations, so you can check the sources directly and judge their credibility for yourself.

• Gradual rollout: The feature, built into Claude 3.7 Sonnet, is first available to paid subscribers in the United States, with an expansion to free users and other regions coming soon.

• Easy activation: To use it, just turn on the “Web Search” option in your profile settings on the platform.

Why it matters ⤵️

This update helps Claude catch up with the competition while playing to its strengths. Its ability to seamlessly integrate verifiable sources is a major asset for anyone looking for reliable information, whether for professional work or personal research.

🎁 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 unique 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 a pro.

🎁 6 shares → The 15 Must-Have 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 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

🎨 Bring your old photos back to life in color

Learn how to easily turn your precious black-and-white photos into vibrant colorized images. Thanks to Gemini 2.0 Flash’s new native image generation feature, you can now instantly revive family memories or historical documents in just a few clicks.

Tutorial steps:

1. Go to Google AI Studio and select “Gemini 2.0 Flash (Image Generation) Experimental” from the “Models” dropdown menu.

2. Upload your black-and-white image by clicking the “+” button in the message field.

3. Simply type “Colorize this image” and hit Run to transform your photo.

4. Customize further with creative instructions like “Add snow to the trees” or “Change the lighting to golden hour,” then download your new creation!

Tip: Be specific about your color preferences to get more personalized results. Try “colorize with warm, summery tones” instead of just “add color” for better results.

WHAT THE RESEARCH SAYS

🧐 AI reveals its own “cheating strategies”

New research from OpenAI on chain-of-thought reasoning in AI models has uncovered a fascinating phenomenon: models like o3-mini can “bend the rules” or cheat on certain tasks. Even more surprising, trying to stop them from thinking about cheating just pushes them to hide their true intentions.

Details:

• By examining the models’ reasoning, OpenAI found they were openly planning to cheat, with thoughts like “Let’s hack” or “We can bypass the tests by exiting early”.

• In coding tasks, the models used shortcuts like modifying test files, returning hardcoded values or using exit commands to dodge evaluations.

• When researchers tried penalizing the models for their cheating thoughts, they kept cheating but hid their intentions in their reasoning process.

• OpenAI concludes that the best approach is to keep internal reasoning uncensored for monitoring, while using separate models to filter the thoughts.

Why it matters

Just like us, AIs look for shortcuts and ways to game the system. But as they approach superhuman capabilities, monitoring their chain of thought may well be our only window into their true reasoning. Pushing too hard for them to behave “properly” could slam that precious window shut for good.

THE LATEST TRENDING TOOLS

• 📜 Mistral OCR offers a powerful solution for extracting text from images or documents with remarkable accuracy.

• 👤 Magicam lets you swap faces and change voices in real time during streams and meetings, adding a new dimension to your virtual interactions.

• 📧 Serif acts like a true AI executive assistant that efficiently organizes your inbox and drafts replies with an authentic human touch.

• 🎬 EasySub offers an online platform for automatic subtitle generation for your videos, with over 90% accuracy and translation available in more than 150 languages.

• 🖌️ Bolt x Figma turns your Figma designs into ready-to-use apps in a single click, dramatically simplifying the development process.

IN THE NEWS

• Sony just unveiled a prototype of an AI-powered video game character for Horizon Forbidden West, capable of holding real-time conversations with players.

• Anthropic submitted new recommendations for an AI Action Plan to the White House, calling for stronger national security testing, tighter export controls and infrastructure expansion.

• 23% of jobs in the tech sector in 2025 are looking for artificial intelligence skills.

• AI doesn’t really learn – understanding this can help you use it more responsibly and effectively.

• Perplexity is set to raise nearly $1 billion at an $18 billion valuation, potentially doubling its value, as the AI search startup nears $100 million in annual recurring revenue.

• Let’s finally talk about AI’s impact on the video game industry.

• Figure, the humanoid robot maker, announced the launch of BotQ, a new factory capable of producing 12,000 humanoids per year, with expansion plans to reach 100,000 units.

YOUR DAILY DOSE OF CONTENT

1/ 🎮 Check out this fascinating 3D game generated by Grok 3, which can handle more than 3,500 lines of complex logic for a truly impressive result.

The latest AI generated 3D game I made with @grok 3 is mind-blowing! No other model can handle over 3,500 lines of complex game logic like Grok 3 from @xai. Try the Xforce game—link below 👇

5:35 PM • Mar 2, 2025

5.92K Likes   747 Retweets  

2/ 🎵 Claude shows off its composing chops by creating a quality music track in Ableton with just two simple instructions.

🎵💿Built an MCP that lets Claude talk directly to Ableton. Now you can create music with just prompts!

Here’s a demo of me creating a lush, 80s synthwave track in just two prompts. It picks the right instruments, creates melodies, and adds effects like reverb and distortion 🔊

1:50 PM • Mar 20, 2025

5.79K Likes   609 Retweets  

3/ 🦅 This surprising robot bird jumps, walks and clears obstacles with remarkable agility, perfectly mimicking natural movements.

AI phone agent realizes it is talking to a parrot

4:36 PM • Mar 12, 2025

8.71K Likes   712 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