Hey AI Enthusiasts!

This week’s roundup is packed with some of the most exciting breakthroughs in AI. From Google’s upgraded video model and MIT’s self-evolving system to a Google model that helped discover a new cancer treatment.

Let’s dive in!

In today’s insights:

🎬 Google releases Veo 3.1 with major video + audio upgrades
🧠 MIT unveils a self-evolving AI that rewrites its own code
💊 Google’s Cell2Sentence 27B helps predict new cancer treatments

_{Read time: 5 minutes.}

Google
🎬 Google Unleashes Veo 3.1: AI Video Takes a Cinematic Leap

The AI Field: Google has released Veo 3.1, an upgraded AI video generation model that delivers significantly improved audio generation, better prompt adherence, and enhanced realism when converting images and text into videos. The model can now generate videos up to 60 seconds long at 1080p resolution with synchronized native audio including dialogue, sound effects, and ambient soundscapes.

Details:

• Veo 3.1 integrates deeply with Google's Flow video editor, adding native audio generation to features like "Frames to Video" (generate content between two keyframes), "Ingredients to Video" (use reference images for characters and styles), and "Extend" (seamlessly continue existing clips)

• New editing capabilities include "Insert" for adding objects that blend naturally into scenes with proper lighting and shadows, plus an upcoming "Remove" feature to erase unwanted elements

• The model delivers superior character consistency across frames, more realistic textures, and better control over camera movements and cinematic styles compared to its predecessor Veo 3

• Available now through the Gemini API (approximately $0.40/second for Standard, $0.15/second for Fast), Vertex AI for enterprises, the Gemini app, and third-party platforms like Freepik and ImagineArt

Why This Matters: This release marks a fundamental shift from basic AI video generation to precise, iterative creative control—pushing the technology beyond novelty clips toward viable professional production tools. The integration of synchronized audio with extended duration positions Veo 3.1 as a direct competitor to OpenAI's Sora 2, potentially democratizing video creation for marketers, educators, and content creators who previously needed expensive equipment and technical expertise. However, the increased realism also amplifies concerns about deepfakes and misinformation, making Google's watermarking and detection efforts critical for maintaining trust in digital media.

👉 Read more here

How can AI power your income?

Ready to transform artificial intelligence from a buzzword into your personal revenue generator

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

• A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential

• Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background

• Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

MIT
🧠 MIT's Self-Evolving AI: Models That Rewrite Their Own Code

The AI Field: MIT researchers have developed SEAL (Self-Adapting Language Models), a framework that enables AI models to autonomously generate their own training data and update their internal parameters without human intervention. The system achieved a 40% boost in factual recall and outperformed GPT-4.1 using self-generated data.

Details:

• SEAL uses a dual-loop process where the model reads new information, creates "self-edits" for updating itself, applies updates through fine-tuning, then uses reinforcement learning to reward successful improvements

• In tests, SEAL improved question-answering accuracy from 32.7% to 47.0% after learning new facts, and achieved 72.5% success on few-shot learning tasks compared to 0% for traditional in-context learning

• The framework is now open-sourced under an MIT License and was presented at NeurIPS 2025

Why This Matters: This breakthrough challenges the core limitation of current AI that models remain static after deployment. SEAL enables continuous learning without manual retraining, particularly valuable for enterprise systems that need to process new information constantly. However, the research revealed a critical weakness: catastrophic forgetting, where new learning can overwrite previous knowledge. If solved, self-adapting models could transform AI development from a labor-intensive process into an autonomous evolutionary system.

👉 Read more here

Google
💊 AI Predicted a New Cancer Treatment and Then Proved It in the Lab

The AI Field: Google's Cell2Sentence-Scale model analyzed over 4,000 drugs and discovered that combining silmitasertib with low-dose interferon makes "cold" tumors visible to the immune system, a prediction that was then confirmed in human cell experiments.

Details:

• Lab tests showed the drug combination produced a 50% increase in antigen presentation, potentially making tumors more responsive to immunotherapy

• This connection had never been reported in scientific literature before—the AI generated a genuinely novel hypothesis, not just patterns from existing research

• The 27 billion parameter model is built on Google's Gemma and is now open-sourced for researchers

Why This Matters: For the first time, an AI didn't just analyze existing cancer research, it predicted an entirely new treatment approach and got it right. This could transform drug discovery from a decades-long process of trial and error into targeted AI-guided breakthroughs, potentially accelerating life-saving therapies from lab to clinic.


👉 Read more here

💥 OpenAI x AMD: $Billion GPU Deal
OpenAI has struck a massive deal with AMD to buy 6 gigawatts of Instinct GPUs, starting in late 2026. The deal could give OpenAI up to a 10% stake in AMD. AMD shares jumped 24%, adding $63B in market value.

🌍 Anthropic x Deloitte Partnership
Anthropic just signed its biggest corporate deal ever: Deloitte will deploy Claude to 470,000 employees in 150 countries. They’ll also launch a Claude certification program for 15,000 professionals and a new Center of Excellence.

⚠️ Yoshua Bengio’s AI Warning
AI pioneer Yoshua Bengio warns humanity could face extinction within a decade if AI systems evolve without safeguards — suggesting they might soon prioritize self-preservation over human goals.

🇪🇺 Europe Doubles Down on AI Sovereignty
European leaders are boosting investments in “AI sovereignty,” focusing on defense command systems and empowering local champions like Mistral and Helsing to strengthen Europe’s AI independence.

Golden Nuggets

🎬 Google releases Veo 3.1 with major video + audio upgrades
Veo 3.1 delivers smoother motion, native sound, and lifelike realism, turning text prompts into cinematic clips.

🧠 MIT unveils a self-evolving AI that rewrites its own code
MIT’s SEAL framework lets AI models improve themselves using self-generated data, no human retraining needed.

💊 Google’s Cell2Sentence 27B helps predict new cancer treatments
Google’s model found drug combinations that made tumors visible to the immune system, later confirmed in lab tests.

Until next time!
Olle | Founder of The AI Field