Descry AI

Descry AI is an 8-week AI safety technical talent incubator designed for exceptional high school students in the Global South. By engaging potential student researchers, they are provided an accessible and rigorous entry point into AI safety research. Through hands-on mentorship, technical training, and structured research opportunities, participants’ skills will be developed. Curated by graduates of the Technical Alignment Research Accelerator (TARA) and published high school students, it allows for the further development of local AI safety ecosystems. Ultimately, such endeavors will contribute to current risk-mitigation efforts in AI research, especially for underrepresented sectors in the Global South.

Within this program, participants may choose between two tracks.

1. The Mentor Track is for students who would like to be mentored under a specific mentor’s research agenda. They may further their work into a specific research interest under experienced AI safety mentors. 

2. Alternatively, the Open Track allows students to develop independent research practice through paper replication and exploratory projects of their chosen research interest. A generalist research mentor will oversee the development of their research project and assist in guiding the student.

Students will have the option to pursue any of the following five research interests.

1. Low-Resource Sociocultural Linguistics: Under this theme, research delves into AI safety and how it survives specific sociocultural contexts, especially in the Global South. 

2. Information and Bias: This theme relates to how AI models represent knowledge and truth, ruminating on the risks of misinformation and/or disinformation. Practically, it can be applied to the intersecting realms of politics and social bias.

3. Interpretability for Biology and Medicine: Works under this theme sit where AI safety meets the life sciences. This intersection allows students to explore biosecurity as a field, identify implicated biological risks, and develop tools that enable safer deployment operations.

4. Safety at the Edge: This theme investigates the robustness of AI safety mechanisms and evaluations. This is done especially in the context and conditions of the Global South, where local limitations may amount to a number of other risks.

5. AI for High-Stakes Local Problems: AI is increasingly being applied to local issues such as healthcare, disaster preparedness, agriculture, education, and a multitude of other public services. This theme focuses on how AI models may be utilized in this respect and their capacity for real-world deployment.