What if machine learning wasn’t chaos but chemistry? MIT’s latest framework unifies decades of algorithms under one equation, predicting the future of AI like Mendeleev with models. Meanwhile, two students with zero funding just built a TTS model that outperforms the pros. Grok’s new vision feature adds real-world sight. And OpenAI wants Chrome, yes, that Chrome, if regulators pry it loose from Google.

• MIT builds a predictive map of ML’s hidden structure
• Dorm-room TTS model topples ElevenLabs in tests
• Grok now answers questions through your camera
• OpenAI may swoop in if Google is forced to sell Chrome

Special highlight from out Network

🌀 Supercharge Your Developers: AI plugins that do the heavy lifting

Tech leaders, are your devs really working at full power?

In today’s fast-moving landscape, AI-powered tools aren’t just a nice-to-have. They’re the key to building better software, faster.

Join mindit.io for a sharp 45-minute webinar:

"Enhancing Developer Productivity: The Power of AI Plugins".

You will also be automatically registered for a GIVEAWAY where you can win 3 AI & Automation for Business Workshops (duration 2h) - hands-on sessions on integrating AI and automation to boost efficiency.

Led by Alexandra Arsene (Technical Lead) and Anca Iordanescu (VP of Engineering, IKEA - Store of the Future), this session dives into how top teams are using AI plugins to:

✅ Accelerate coding
✅ Improve testing
✅ Ship higher-quality products

You’ll get real-world examples, implementation strategies that won’t disrupt your flow, and tips for balancing AI with human insight.

Want your dev team to build faster without burning out?
This session’s for you.

🗓 April 29 at 14:00 CET

MIT researchers have created a unifying framework that organizes over 20 classical machine learning algorithms using a single equation, forming a “periodic table” that reveals deep connections between them and helps scientists design entirely new models.

Key developments:

• A single equation ties together diverse algorithms. The framework, called Information Contrastive Learning (I-Con), shows how machine learning methods approximate relationships between real data points using a shared mathematical structure.

• Discovery began by accident. Lead author Shaden Alshammari, while researching clustering, found it could be mathematically reframed in the same way as contrastive learning, sparking the idea for the table.

• Existing algorithms reclassified. I-Con includes everything from spam detectors to deep learning systems powering LLMs, and categorizes them by how they learn from data and approximate relationships.

• Hidden gaps hint at future models. Like the chemical periodic table, I-Con predicts the existence of new algorithms in currently unfilled slots. The team already created one that improved image classification performance by 8 percent over top methods.

• Cross-pollination of ideas. The team applied a contrastive learning debiasing technique to clustering and saw improved accuracy, demonstrating how the table enables method transfer.

• Framework is expandable. Rows and columns can be added to reflect new types of data relationships, giving researchers a flexible map for algorithm discovery.

The paper was co-authored by John Hershey (Google AI), Axel Feldmann and William Freeman (MIT), and Mark Hamilton (Microsoft), and will be presented at the International Conference on Learning Representations. As Alshammari puts it, the metaphor is real. Machine learning is no longer just guesswork but a structured space ripe for exploration.

Korean startup Nari Labs introduced Dia, an open-source text-to-speech model that outperforms industry leaders like ElevenLabs and Sesame. The model was developed by two undergrad techies with zero funding, setting a new standard for TTS innovation.

How they did it:

• Nari Labs launched Dia, a 1.6B parameter model designed to handle emotional tones, speaker tags, and nonverbal cues like laughter, coughing, and screams.

• The project was inspired by Google’s NotebookLM, with Nari using Google’s TPU Research Cloud for compute access.

• Side-by-side tests show Dia surpasses ElevenLabs Studio and Sesame CSM‑1B in expressiveness, timing, and nonverbal handling.

• Founder Toby Kim revealed plans for a consumer app centered on social content creation and remixing based on Dia’s capabilities.

Dia’s launch is proof that even with zero funding, and minimal experience, innovation is within reach. With AI resources more accessible than ever, it’s clear that the barriers to groundbreaking advancements are lower than most think.

xAI has launched Grok Vision, a new feature that gives its Grok chatbot real-time visual capabilities. Available on iOS, the update allows users to point their phone’s camera at objects and get instant answers, adding another layer of multimodal interaction to the rapidly evolving assistant.

What’s Changing:

• Grok Vision lets users interact through the camera. iOS users can now ask Grok questions about anything in view, from product labels to street signs to printed documents, with the chatbot offering live analysis and contextual information.

• Vision is only live on iOS. The Grok app for Android does not yet support the camera feature, though it’s expected in future updates.

• Grok’s voice mode gets more powerful. In addition to visual input, xAI added multilingual audio and real-time search capabilities to Grok’s voice assistant, offering a more fluid and dynamic experience.

• Some features are paywalled. Android users can access voice upgrades only through the $30-per-month SuperGrok subscription tier.

Earlier this month, xAI added memory recall and a canvas-like interface for building documents and apps, showing continued momentum in feature development. With Grok Vision, xAI is catching up to competitors like Google Gemini and ChatGPT in real-world perception.

During Google’s antitrust trial, ChatGPT chief Nick Turley said OpenAI would be interested in acquiring Chrome if it became available. The move would align with OpenAI’s vision of building an “AI-first browser” and reshape how users interact with the web.

What happened:

• OpenAI exec Nick Turley said in court the company would buy Chrome to offer a “really incredible experience” and showcase an AI-native browser.

• The comment came amid DOJ efforts to determine which parts of Google’s business must be modified or divested after a monopoly ruling.

• The DOJ has proposed forcing Google to divest Chrome as a possible remedy.

• OpenAI has already hired ex-Chrome leads Ben Goodger and Darin Fisher, hinting at browser ambitions.

Browsers are the gateway to the weband owning one could give OpenAI a direct channel for deploying its AI models at scale. If regulators force Google to sell, it could trigger a seismic shift in who controls the future of online search.

🚀 Boost your business with us—advertise where 10M+ AI leaders engage

🌟 Sign up for the first AI Hub in the world.

📲 Our Socials