🕵️ Anthropic CEO’s Behind-the-Scenes Adviser Is His Wife (1)

Welcome back to AI Researches! 👋

This week, Anthropic leads with Claude Fable 5.1, a stronger model for coding, long technical tasks, and agent-style work. It performs better on hard benchmarks, checks its own output more often, and could make Claude workflows cheaper through lower cache-read costs.

The rest of the issue moves across AI in medicine, interfaces, hardware, and productivity. We cover AI-powered ECGs, live AI-assisted brain surgery, Runway’s real-time generated interfaces, Anthropic’s machine-control standard, Claude and Gemini tutorials, expert predictions, useful tools, and more.

Let’s dive in. 

In today’s AI Researches:

• 💻 Anthropic launches Claude Fable 5.1 with a big coding upgrade

• ❤️ AI spots hidden heart disease from a routine ECG

• 🎬 Runway introduces real-time generated video interfaces

• 🧠 UK surgeons perform the first live AI-assisted brain surgery

• 🤖 Anthropic creates a standard for AI agents to control real-world machines

• 🗣️ Learn how to build a website prototype by talking to Claude

• 📊 AI tip: Turn Google Sheets into interactive dashboards with Gemini

• 🔮 Wild predictions from Sam Altman and Bill Gates

• 🛠️ Best performing AI tools of the week

• 📰 More AI news and developments

Read time: 5 minutes

 NEW 

💻 Anthropic Launches Claude Fable 5.1 With a Big Coding Upgrade

AI Researches: Anthropic launched Claude Fable 5.1, a stronger AI model for coding and knowledge work that performs much better on long tasks, checks its own work more often, and costs less to run in many cases.

The Details

• Fable 5.1 scored 52.6% on Terminal-Bench-Science, more than double Fable 5’s 24.7%, showing a big jump on harder technical tasks.

• It also leads CursorBench 3.2 at 73.4%, with stronger agent-style coding where the AI can work longer and verify its own output.

• Anthropic kept the same base price while cutting cache-read costs by 75%, which can lower typical workloads by around 25% and agent-heavy work by up to 45%.

• Fable 5.1 is now available in Claude Code, the Claude Platform, and for Pro, Max, Team, and Enterprise users, with developers already using it for games, design, and longer coding jobs.ip structure.

 iMajor AI News that Happened in this week i

Imperial College London researchers just developed an AI system that can spot hidden heart disease from a routine ECG in seconds.
Trained on millions of ECGs, it can flag heart failure and valve disease that may be invisible to doctors, with NHS trials now moving forward.

Runway introduced Solaris, its experiment in turning interfaces into real-time generated video.
Users can interact with generated screens through clicks and drags while the AI continuously decides and renders what should happen next.

UK surgeons just performed the first live AI-assisted brain surgery.
The UCL-developed system watched the surgical video in real time and highlighted critical blood vessels and optic nerves while doctors removed a brain tumour.

Anthropic introduced a standard for connecting AI agents to real-world machines.
The Model Hardware Standard lets agents learn and operate equipment like microscopes, robotic arms, and lab instruments without weeks of custom integration.

A powerful mystery model just showed up for free on OpenRouter.
Ox Alpha is built for coding and long-running agent tasks, and its strong benchmark results have sparked speculation that it could be an unreleased model from Zhipu AI.

Slack launched a new way to build software collaboratively with AI agents.
Slack Code creates shared coding rooms where agents write software while teammates watch, steer the work, preview changes, and approve deployments.

🗣️ Build a website prototype by talking to Claude

1. Open Claude Voice
Start a new Claude chat and click the voice icon. Voice mode works on Claude Desktop, web, and mobile.

2. Explain the website
Describe the audience, goal, pages, style, colors, content, and any design references.

3. Ask Claude to interview you
Prompt: “Ask me questions until you have a complete website plan and brand direction.”

4. Create the build plan
At the end of the voice chat, ask Claude to write the full website structure, sections, copy direction, and design notes.

5. Build the prototype
Switch to text mode, upload photos or references, and ask Claude to build a static site artifact from the plan. Anthropic’s artifacts guide covers building and sharing prototypes in Claude.

Pro tip: Use the words static site for portfolios, landing pages, and mockups to keep the build simple.

Tutorial link: Claude Voice Mode official guide

Lady Gaga and Michael Polansky’s startup Outer Bio built Yuna, a system that keeps human skin alive for up to four weeks and uses the test data to train AI. The AI then helps find promising skincare compounds much faster, cutting the search for new candidates from many months to about six weeks.

Chinese state-linked hacking groups have reportedly more than doubled their attacks after adding open-source AI models like DeepSeek to their tools. Researchers say the hackers use it to write attack code, scan targets, and break into systems faster because it is cheap, powerful, and has fewer safety limits.

Researchers at the University of Northampton used AI to study breast cancer samples cell by cell and found hidden patterns linked to how aggressive a tumor may be. The findings could eventually help doctors better predict outcomes and choose more personalized treatments.

📊 Turn Google Sheets into Interactive Dashboards with Gemini

1.Open your spreadsheet
Open a project tracker, content calendar, budget, or other dataset in Google Sheets, then open the Ask Gemini side panel and select Create canvas.

2. Describe the dashboard you want
Try a prompt like: “Turn this spreadsheet into an interactive dashboard with charts, key metrics, and filters.”

3. Let Gemini build it
Gemini turns your spreadsheet data into an interactive visual mini-app while keeping it connected to the original Sheet.

4. Refine and share it
Ask Gemini to change the layout, charts, or functionality until it looks right, then share the Sheet with others like normal.

🔗 Learn more→

Sam Altman says OpenAI expects to have a model that meets his personal definition of AGI internally by the end of the year, while research chief Mark Chen believes the company is already about 80% of the way there. OpenAI also says its Astra model can handle roughly a week of research work on its own and may be the first system to meaningfully create new knowledge.

Bill Gates says the world still has “no plan” for how to handle AI’s impact on jobs, and has suggested taxing AI tokens and robots while reserving some roles specifically for humans.

• 🎥 LazyKiwi — Transforms photos into social-ready AI videos using hundreds of effects and templates built for Reels, TikTok, and Shorts.

• 📄 Huntr — Tailors your resume to specific jobs by matching relevant keywords, responsibilities, and qualifications while keeping your entire job search organized.

• ⚖️ Finest — Routes each AI request to the lowest-cost model proven to meet your quality requirements, helping cut API spend without blindly downgrading performance.

• 💼 Jobseeker — Helps create and tailor CVs, write targeted cover letters, discover openings, and keep every job application organized in one place.

• 🎬 CraftStory — Turns a single photo and script into a lifelike AI presenter with consistent identity, natural lip-sync, and videos up to 30 minutes long.

• 🔎 krucbl — Stress-tests startup ideas across 14 different angles, using live market research to uncover competitors, weak assumptions, and the smartest next move.

Anthropic just tested Claude agents doing AI safety research largely on their own.
The agents developed fixes across 10 failure modes like deception, sycophancy, and jailbreaks, outperforming experienced human researchers in Anthropic’s benchmark setup

Bernie Sanders just published a new op-ed calling for an international pause on more powerful AI models.
He argues that recent security incidents, job risks, and growing loss of control make coordinated action between countries like the U.S. and China increasingly urgent.

OpenAI just announced it will remove its models from Cursor after the SpaceX acquisition. Access is expected to end on November 12, with OpenAI citing concerns over Elon Musk-owned companies’ past contract violations.

SpaceXAI and Cursor launched Grok Bot, a new AI teammate platform.
The beta gives Grok-powered agents their own computers, lets them work autonomously across apps, and allows multiple bots to coordinate with each other.

OpenAI says its first custom AI chip can outperform Nvidia on speed and efficiency.
Jalapeño is built for running AI models, not training them, and reportedly delivers up to 3.6× faster inference with up to 1.9× better work per watt.

SpaceX is building its orbital AI data centers around Nvidia’s Vera Rubin platform.
The company plans to use a lighter, space-optimized version of Nvidia’s NVL72 system for its Starmind infrastructure, with deployment targeted for late next year.

ChatGPT can now work inside Apple Messages. OpenAI’s new plugin gives its apps access to conversations so users can find messages, read threads, and send replies with AI.

Thomson Reuters just built its own legal AI model. The company reworked Alibaba’s open-source Qwen and trained it on decades of proprietary legal and professional content, spending about $40M on the effort.