Welcome, AI Enthusiasts!

Google is taking the lead in supporting AI startups with its new AI Futures Fund, while major advancements in AI’s role in healthcare are being made with tools like FaceAge. Meanwhile, Sakana AI is introducing brain-inspired model and Microsoft is making new developments that signal greater autonomy in its AI systems.

Models are no longer just trained, they're teaching themselves.

In today’s Generative AI Newsletter:

• Google backs startups with DeepMind tools and cash, no cohort required
• AI reads faces to predict cancer survival, outperforming doctors
• Sakana unveils CTMs, brain-inspired models that reason step by step
• Microsoft Build preview hints at full-stack Copilot with custom models and chips

🚀 Google Launches AI Futures Fund for Startups

Google has launched the AI Futures Fund, a new program to support startups building with its latest AI models. Run outside of any cohort system, the fund operates on a rolling basis and offers early access to DeepMind tools, technical mentorship, cloud credits, and potential investment.

Key points:

• Startups gain access to DeepMind models, including unreleased research tools

• Support includes technical guidance from DeepMind and Google Labs engineers

• Google Cloud credits are available to support scaling

• Investment is available from seed to late stage, based on startup needs

• Early participants include Viggle (AI motion memes) and Toonsutra (AI webtoons)

This is part of Google's broader strategy to embed itself in the AI startup landscape. The fund follows recent initiatives including a $20 million AI research grant, a $120 million global AI education push, and a nonprofit AI accelerator. With DeepMind at the center, Google is turning its internal breakthroughs into an external advantage.

📸 AI Reads Your Face to Predict Cancer Survival

Mass General Brigham scientists have introduced FaceAge, an AI model that predicts biological age from a single facial photo and uses it to estimate cancer survival. Published in The Lancet Digital Health, the study found that FaceAge outperforms standard clinical tools, without needing lab tests or scans.

Key developments:

• Trained on 58,000 healthy faces and tested on 6,196 cancer patients across the US and Europe

• Outperformed doctors in predicting survival across curative, thoracic, and palliative cancer groups

• Spotted a 5-year average gap between biological and actual age in cancer patients, linked to poorer outcomes

• Improved survival accuracy when used alongside clinical records

• Revealed genetic links to senescence pathways that traditional age metrics missed

FaceAge is not yet ready for hospitals, but it hints at a future where fast, image-based tools help doctors make complex calls. With healthcare systems under strain, non-invasive models like this could scale rapidly especially in critical areas like cancer care.

🧠 Sakana’s New Model Thinks Like a Brain

Sakana AI has introduced Continuous Thought Machines (CTMs), a new kind of AI model that mimics how the brain reasons through problems. Instead of making instant decisions like most AI, CTMs “think” through problems step-by-step, using internal dynamics that evolve with time.

Key innovations:

• Time-aware neurons: CTMs don't just consider current input, but track their own activity history to guide decisions.

• Neuron synchronization: Inspired by real brains, CTMs coordinate neuron activity over time, unlocking richer, more adaptive behavior.

• Emergent reasoning: In maze-solving tasks, CTMs learn to trace paths naturally without explicit programming. For image recognition, they move their attention across key features in human-like patterns.

While most AI ignores timing and neuron interaction, Sakana’s CTM introduces a temporal reasoning layer that’s closer to how biological intelligence works. The result is a more interpretable, flexible, and efficient model, one that could close the gap between artificial and natural cognition.

Bonus: You can literally watch CTMs "think." The decision-making is fully observable, offering transparency rarely seen in deep learning systems.

🔮 What to Expect at Microsoft Build 2025

Microsoft’s flagship developer event kicks off May 19, and it could mark a turning point in the company’s AI roadmap. The company appears to be moving toward a future where its own models, chips, and agentic systems play a central role, with less reliance on OpenAI.

What to expect:

• Copilot autonomy: Leaked features like an “Action” button suggest AI that can execute PC tasks unprompted

• Model diversification: Microsoft is reportedly testing models from Meta, xAI, Anthropic, and DeepSeek inside Copilot

• MAI model family: A new set of in-house models could be revealed, with possible API access

• Maia 2 accelerator: A sequel to its Azure AI chip, rumored to be built by Marvell, may debut

Microsoft is building toward a Copilot that runs on Microsoft chips, powered by Microsoft models, inside Microsoft platforms. If it pulls this off, Build 2025 could mark a decisive pivot from OpenAI collaborator to full-stack AI sovereign.

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

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

📲 Our Socials