SunBrief#51: DeepMind’s New AI Breakthrough in Cancer Research

Welcome to the SunBrief

Today in SunBrief 🌞

• THIS WEEK: Join the Hidden Barriers of AI ROI Free Event - October 22

• Google’s DeepMind Unveils Breakthrough in AI-Driven Cancer Research

• Anthropic Unveils Claude Haiku 4.5, a Faster and Cheaper AI Model

• Stock Updates

• Amazon’s Ring Partners With Flock to Expand AI Surveillance Access

• AI Highlights of the Week

• Too Important to Miss

THIS WEEK: Join the Hidden Barriers of AI ROI Free Event - October 22

Organizations are facing barriers when it comes to getting ROI from AI. Section surveyed 1000s of AI executives to learn what these are.

Spoiler: They're all solvable, and all leadership-led.

Join this free event on October 22 to hear Section's proprietary insights on the biggest AI friction points and how to get AI initiatives on the right track, based on real experiences from 10,000s of knowledge workers and heads of AI.

Google’s DeepMind Unveils Breakthrough in AI-Driven Cancer Research

DeepMind Model Generates and Validates a New Cancer Treatment Hypothesis

DeepMind’s new AI discovered a drug combo that helps the immune system better detect tumors—a milestone for AI in cancer research.

Key Points:

• AI-Driven Discovery: The 27-billion-parameter model, built on Google’s Gemma framework, generated a new cancer-therapy hypothesis that was later confirmed in human cell experiments.

• Novel Mechanism: DeepMind found that the drug silmitasertib (CX-4945) can enhance antigen presentation—making tumors more “visible” to the immune system—when paired with low-dose interferon.

• Experimental Validation: Tests on neuroendocrine tumor cells showed a 50% increase in immune signaling, confirming the model’s prediction.

• Scientific Significance: Researchers say this demonstrates that scaled biological AI can generate new scientific ideas rather than just optimize existing ones.

Why It Matters:

DeepMind’s breakthrough marks a shift for biological AI, proving it can generate and test new scientific ideas, advancing cancer research and drug discovery.

Anthropic Unveils Claude Haiku 4.5, a Faster and Cheaper AI Model

Haiku 4.5 Delivers Near-Frontier Coding Performance at One-Third the Cost

Anthropic’s new Claude Haiku 4.5 matches Sonnet 4’s coding power while running twice as fast at one-third the cost.

Key Points:

• Performance Leap: Haiku 4.5 offers near-frontier coding ability and even surpasses Sonnet 4 in some computer-use tasks, making it ideal for real-time assistants, customer support, and pair programming.

• Cost and Speed: Available at $1/$5 per million input and output tokens, Haiku 4.5 enables developers to run high-performance AI applications at a fraction of previous costs.

• Multi-Agent Collaboration: Users can pair Sonnet 4.5 and multiple Haiku 4.5 instances to break complex tasks into coordinated subtasks for faster, parallel execution.

• Broad Availability: Haiku 4.5 is accessible on Claude Code, Anthropic’s API, Amazon Bedrock, and Google Vertex AI, serving as a drop-in replacement for Haiku 3.5 and Sonnet 4.

Why It Matters:

Claude Haiku 4.5 delivers top-tier AI performance with greater speed and efficiency, setting a new standard for affordable, reliable AI tools.

Stock Updates

Amazon’s Ring Partners With Flock to Expand AI Surveillance Access

Collaboration Raises Privacy Concerns as Law Enforcement Gains Wider Footage Access

Amazon’s Ring is partnering with Flock, letting law enforcement request footage from Ring users through Flock’s AI-powered surveillance network.

Key Points:

• Expanded Reach: The partnership could link millions of Ring cameras to Flock’s AI surveillance network, allowing police to scan license plates and identify people or vehicles.

• Privacy Concerns: Experts warn the collaboration may worsen racial bias and expand government surveillance powers.

• Federal Access: Agencies including ICE, the Secret Service, and the U.S. Navy already use Flock’s data, widening federal reach.

• Security Issues: Ring’s history of privacy lapses led to a $5.8M FTC fine in 2023 for improper access to user videos.

Why It Matters:

The Ring–Flock deal expands AI surveillance in the U.S., sparking new concerns over privacy and civil liberties.

AI Highlights of the Week

• Google Begins Quiet Rollout of Gemini 3.0 Pro

  Google is stealthily rolling out Gemini 3.0 Pro, its smartest AI yet, ahead of an official launch. Users spotted upgrades on the Gemini site and AI Studio.

  The model shows big improvements in coding tasks and will power Gemini Enterprise, Google's push for a unified AI platform for businesses.

• Kayak Adds 'AI Mode' for Smarter Travel Planning

  Kayak’s new AI Mode, powered by ChatGPT, lets users ask travel questions, explore trip ideas, and book flights, hotels, and cars.

  It's live in the U.S. on web and will soon expand to more regions, with voice support coming soon.

• Google Launches Veo 3.1 with Realistic Video Upgrades

  Google’s new Veo 3.1 brings richer audio, lifelike textures, and improved storytelling for video generation via the Gemini app, API, and Vertex AI.

  The Flow tool now supports audio, scene extension, and realistic edits—soon letting users remove objects with seamless background reconstruction.

• Amazon Previews Modular Nuclear Plant in Washington

  Amazon has shared a first look at the Cascade Advanced Energy Facility, a compact modular nuclear plant near Richland, WA, developed with Energy Northwest.

  It’s expected to produce 960 MW, power 770,000 homes, and create 1,000+ jobs, with construction starting later this decade.

Too Important to Miss

Spotify, labels launch “responsible” AI music lab

Apple is building a touchscreen MacBook Pro for the M6 era

New York enacts first statewide ban on AI rent-fixing software

Google’s Search AI Mode expands to 200+ regions, 35+ languages

Anthropic debuts Claude “Skills” to make its AI actually useful at work

Last Week’s Poll Result

1. Do you think Tesla can overcome the design issues with its Optimus humanoid robots?

   Yes: 37.78%, Maybe: 33.33%, No: 28.89%

2. Have you tried OpenAI’s new video-generation app, Sora?

   Yes: 31.58%, No: 21.05%, Maybe: 47.37%.

3. Are you concerned about the environmental impact of Starlink satellites burning up in the atmosphere?

   Yes: 43.48%, No: 28.26%, Maybe: 28.26%.

Feedback

Know someone who may be interested?