kannan

Rebuilding Four Small Projects

The author redesigned four open-source projects—DonumAI, MittentisAi, Learn with DonumAI, and EZTime—to simplify their architectures and make them easy to document with single diag…

The Rate at Which We Erase the Future

The author argues that applying a constant social discount rate to environmental policy is mathematically and ethically flawed, as it tends to undervalue the long-term impacts of c…

What Should Students Be Taught on Using AI?

Instead of focusing on short-lived prompt engineering skills, the article argues that AI literacy should prioritize teaching students to think critically about how AI systems actua…

Chasing the Next RSA Milestone

By mapping historical RSA breakthroughs against a timeline, the author predicts that the next milestone, RSA-270, could be reached by late this year. However, the recent rapid fact…

Reflections on New York City's AI Moratorium

New York City's one-year moratorium on generative AI for K–8 students highlights the complex tension between the potential benefits of the technology and the institutional need for…

Social Media and Adolescent Well-Being

This article examines how social media's engagement-driven algorithms and lack of transparency pose significant developmental risks to adolescents. While countries like Australia a…

From Semantic Reuse to Auditable Tutoring

The article proposes a framework for scalable AI tutoring that integrates the Crowdsourced Semantic Cache Network (CSCN) with a glass-box learning architecture. This system reduces…

Technical Setup of "Reasoning Over Recall"

The "Reasoning Over Recall" study investigates whether providing Large Language Models (LLMs) with incomplete legal context via RAG—specifically by omitting controlling precedents—…

Reasoning Over Recall

Research indicates that providing AI models with incomplete legal context—such as a statute without its controlling case law—can actually degrade reasoning performance through a ph…

Crowdsourced Semantic Cache Network

The Crowdsourced Semantic Cache Network (CSCN) reduces LLM inference costs by using a shared vector database to intercept semantically similar queries before they require expensive…

Democratizing Personalized Learning using Open Source AI

To solve the "2 Sigma Problem" of personalized instruction, the author argues against the current industry reliance on closed-source AI due to significant risks regarding privacy, …