Sapien Weekly Digest

Hey there, Sapiens! We’re officially in-between the holidays and we’re aiming to end the year strong!

What we’ve been up to

Sapien Team Takeover:
Today, users on our Discord server had the option of chatting with our Chief of Staff, Tyler Koverko, about the upcoming plans for Sapien and what the community can get excited about. We are aiming to run these AMA Style chats on a regular basis to give you all as much insight into the inner workings of Sapien as possible.

A whole lotta planning:

We’re getting ready to hit the ground running in 2026! Our main focus is on making Sapien easier to adopt. Our goal is that anyone, whether they’re a university, an enterprise, a startup, or even a private person that has a need or want to build an AI dataset will be able to rely on Proof of Quality.

🚀 What’s New in the App?

We’ve been working on preparing more tasks for you! We’re aware that this is the community‘s biggest ask and we’re moving all dials to make sure we can guarantee a solid flow of new tasks for you to work on in the coming week.

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. 👩‍🍳 Gastro Tag

7. 👋 Hand Object Annotation

8. 🐞 Bug Hunters

9. 📦 Breakfast Image Bounding Boxes

10. 🚗 Vehicle Model Mapping

Where are we going?

More Sapien Team Takeovers! We got some positive feedback from your end on the test run of the Takeover we had with our Chief of Staff Tyler today! Expect us to expand the scope of these small events.

Christmas! It’s the season yet again, and rest assured we’ll make sure that the Discord will feel festive. Stay tuned for further announcements!

Our Voices in the World

"All the open internet has been fully crawled and trained on. For these models to keep getting better, they have to find new sources of data."

Check out our Co-Founder Trevor Koverko's interview with Cointelegraph talking about how crypto rebuilt his life and how the blockchain democratizes access to AI training.

We’ve also been featured on the AI Chopping Block where our CEO & Co-Founder Rowan Stone spoke on why simple data labelling is falling out of favour and why our Proof of Quality paves the way for quality data at scale.

What else happened in AI?

The World outside of Sapien:

U.S. launches the Genesis Mission:
On November 24, the U.S. administration launched the “Genesis Mission” by executive order under the Department of Energy. The initiative seeks to integrate national laboratories, supercomputers, and AI infrastructure into a coordinated research platform. Its goal is to accelerate scientific discovery in energy, health, and fundamental science. Officials describe the mission as a way to upgrade the tools and workflows available to federally supported researchers. Over a ten year horizon, the program aims to double the productivity of American science and engineering. The long-term impact will depend on governance, data access, and collaboration with the broader research ecosystem. Genesis builds on a recent U.S. AI Action Plan focused on deregulation and accelerating AI deployment across the economy.

Open IMO-Gold Math Model From DeepSeek:
DeepSeek has released DeepSeek-Math-V2, an open source mixture-of-experts model focused on advanced mathematical reasoning.  The model delivered a gold-medal performance at the IMO 2025 contest, with a competition record of a score of 118 out of 120 on the 2024 Putnam exam. The system additionally solved five of six problems on the 2025 IMO, aligning with top human contestant performance. On IMO ProofBench, a benchmark focused on full proof quality rather than short answers, DeepSeek-Math-V2 reached 61.9 percent. This result is close to Google’s Gemini Deep Think, which is tuned specifically for Olympiad proofs, and notably higher than GPT-5’s 20 percent score. DeepSeek attributes the gains to a generator–verifier training setup that rewards logically coherent derivations. In this design, a verifier model assigns confidence scores to individual proof steps instead of judging only final outcomes. By open sourcing a system with this level of performance, DeepSeek aims to make research grade mathematical reasoning more broadly accessible to developers, researchers, and applied domains such as engineering.

AI Detects Dementia via EEG Signals:
A recent study from Örebro University presents EEG based AI systems as a possible tool for early dementia detection. The researchers trained models on resting state EEG data from 88 participants, including cases of Alzheimer disease, frontotemporal dementia, and healthy volunteers. Their main deep learning framework combines temporal convolutional and long short term memory layers to model time varying spectral features. Using explainable AI methods, the team generates visual maps that highlight which brain wave bands and scalp locations most strongly affect each classification. In cross validation experiments, the network reached about 80 percent accuracy when separating Alzheimer disease, frontotemporal dementia, and healthy status. Performance was even higher on simpler tasks that only distinguished dementia from no dementia, where accuracy exceeded 97 percent on the same dataset. The authors suggest that such lightweight and privacy preserving models could eventually run on portable equipment in clinics or even homes.

See you next week!