Sapien Weekly Digest

Hey there, Sapiens!

We’re already halfway through November, can you believe it? As we’re moving closer to Thanksgiving, here's the week in Sapien!

What we’ve been up to:

Town Hall Meeting:
Today, our CEO Rowan Stone and our Head of Product, Lauren Mushro (read her Newsletter HERE), joined us on stage to talk about what we have been building since TGE, and where we’re expected to go from here.

If you’ve missed it, no worries! We’ll be posting clips and quotes periodically. It’s our goal to have at least one major Town Hall meeting like this per month, so make sure to set your reminders for next time! We’ll make an announcement once we have confirmed a time and date.

Some top level highlights:
We’re expanding our task ecosystem across beginner, intermediate, and advanced annotation types. The training hub will receive new skills assessments, structured learning paths, and an onboarding flow that routes every new user directly into training.

We expect sustained enterprise interest in 3D and 4D data for autonomous vehicles, robotics, and computer vision. Demand is also rising for model benchmarking and minority-language data. The leadership noted that simple annotation tasks will lose value as models improve, while human-in-the-loop reasoning, edge-case identification, and safety-critical interpretation will gain importance.

🚀 What’s New in the App?

Since last week, we’ve updated to Sapien v0.190.0 - Here’s what’s changed:

• Skills test bug fixes

• New Car Model Prompt Matching Module available

• Mobile UI fixes

• Onboarding flow fixes

• A/B testing functionality

• Reputation Score Adjustments

• Tagger Feedback Bug Fixes

• Speech to Text Functionality

• Batching Functionality with Reputation Score

Current task Overview:

1. 🍳 Breakfast Image Upload

2. 🍽 Dinner Image Upload

3. 🧩 Logic Path – Easy / Intermediate / Hard

4. 🌐 Cadena Lógica (Español)

5. 🗣 Njia ya Mantiki (Kiswahili)

6. 🧠 Đường dẫn lô gic (Tiếng Việt)

7. 🧬 Gastro Tag

8. 🐞 Bug Hunters

9. 📦 Breakfast Image Bounding Boxes

10. 🚗Vehicle Model Mapping

New Tasks:

Vehicle Model Mapping is live NOW!

We’ve finally been able to hit the Go! Button on Vehicle Model mapping. For this task, users review the manufacturer and model name provided, search the catalog for the closest match, and assign the correct brand, series, and model to the vehicle presented. This is important so a self-driving vehicle will be able to understand the minute difference between different car models and makes, and improve decision making in edge case scenarios where an ordinarily trained model might guess.

Next week, we’ll also be releasing a new Bounding Box Skills Assessment.This will be your ticket to some of the more complex tasks we’ll be adding in the near future.

Our Voices in the World.

“We believe humans should own their data. We’re all trying to build a future where power is distributed and people are rewarded for their contributions.”

Read our co-founder Trevor Koverko’s detailed interview with CNN on Building a Global Human-Powered AI Workforce Through Blockchain Incentives.

Our own research this week focused predominantly on what it means in detail to build a futureproof regulatory framework around AI, and why the only way we can guarantee expert level knowledge in an AI Model’s output is through making them work smarter, with humans continuously in the loop.

Where are we going?

Sapien Team Takeover. Going forward, we’ll aim to have a team member in the Discord every week for a  structured weekly spotlight that brings the product team closer to the community. Each week a selected team member will lead an asynchronous takeover in Discord to showcase project work, personal reflections, and behind-the-scenes insight.

The featured team member shares short updates, media clips, reflections, or process snapshots throughout the week and engages directly with the community. A running queue of community questions used for AMAs and monthly town halls.

Thanksgiving and Christmas events! After the massive success of the Hallo-Meme contest, we’re aiming to run more events across all our Social Media platforms. Keep an eye out for when we’ll announce Sapiens-Giving. (Working Title.)

In the Pipeline:

Even easier Onboarding. We’re building out ways in which users can build their reputation more reliably and transparently. It is our goal to give everybody the opportunity to use their specific set of skills to benefit the datasets we’re building.

Skills & Tasks. A big focus in what tasks we’ll be delivering to the community will be on Robotics. What makes a robot know how to identify a moving box on an assembly line? You will be the ones to give them the know-how on what to focus on and what they’re seeing.

A new App Design. This one is, truthfully, still a couple months away, but we’re in the process of building a new version of the user facing app, and making it ever easier for the users to find exactly what they want from where they want to find it.

As always, we’re continuously building, and we commit to being as transparent as we can throughout that process!

What else happened in AI? The World outside of Sapien:

Fei-Fei Li's World Labs launches Marble:
World Labs, co-founded by Fei-Fei Li in 2024, has launched Marble, its first commercial “world model,” which generates editable 3D environments from text, images, and video.

Marble is presented with the goal of enabling richer interactions with virtual worlds than those offered by language- or image-only systems. Alongside the launch, Dr. Li has released an essay arguing that the next significant phase of AI progress will depend on spatial intelligence, arguing that the next step for real world AI usage will be models that capture structure, geometry, and dynamics of the physical world. The essay positions Marble as an early realization of this vision, aimed at bridging current generative AI capabilities and future applications in areas such as immersive content creation, simulation, and embodied systems.

A (completely) autonomous Cyberattack:
Anthropic disclosed that a Chinese state-sponsored group hijacked its Claude Code tool to run a largely autonomous cyber espionage campaign against roughly 30 high value organizations. The company detected the activity in mid-September 2025, investigated for about ten days, and then shut down the operation while notifying victims and authorities. Anthropic and several outlets report that Claude performed 80-90 percent of the campaign’s operational workload, with humans mainly approving major decisions and directing which organizations to target.

The attackers bypassed built-in safeguards largely through prompt “role-play,” instructing Claude to behave as if it were supporting legitimate penetration testing and decomposing malicious goals into smaller, ostensibly benign tasks. While presenting the capabilities of AI scaled cyber operations, experts have claimed that the campaign was really an advanced form of automation that still relied on human oversight at key decision points

ElevenLabs launches licensed “Iconic Voice Marketplace” for AI celebrity voices:
ElevenLabs has launched a rights-licensed “Iconic Voice Marketplace” that lets brands and studios legally use AI-generated voices of celebrities and historical figures, based on direct deals with living talent and estates. Actor Michael Caine is the marquee new addition and has formally partnered with ElevenLabs for both the marketplace and the ElevenReader app.

It is an early but important test case for consent-based synthetic performance and posthumous IP in commercial media. The company positions the marketplace as a “performer-first” and consent-based licensing model, brokering contracts between brands and the rights holders of living celebrities or estates, then generating the synthetic voice accordingly. The launch catalog includes 28 curated voices, mixing living celebrities and deceased icons drawn from film, literature, science, and sports.

This raises governance questions about how estates define acceptable uses, where reputational risk sits, and how long posthumous commercialization remains appropriate.

See you next week!