STEM Instruction

Tynker Offers Peek at New LLM-Powered Coding Aid for Students

• By Kristal Kuykendall
• 10/20/23

K–12 game-based coding platform Tynker has developed an AI-powered solution for ages 6–12 called Tynker Copilot that leverages the capabilities of Large Language Models to transform students’ ideas into visual block code for apps and games, the company said in a news release.

Tynker Copilot also can explain block code fragments to student coders as they learn to work with AI in a block coding environment, according to the announcement. The technology is in beta; a general release date has not been set.

Tynker Copilot “empowers children as young as 6 to input text commands like ‘Design a space-themed knock-knock joke’ or ‘Teach me how to build a Fruit-ninja style game’ and receive block code outputs with step-by-step instructions,” the company said. “Moreover, when debugging existing projects, students can submit block-code snippets and benefit from LLM-generated natural language explanations.”

Calling it “the future of coding education,” Krishna Vedati at Tynker parent company BYJU’s, in LinkedIn post, said Tynker Copilot was developed by “fine-tuning Meta’s Llama2-13B model with thousands of lesson examples and kid-created coding projects on a budget of less than $10,000.”

A blog post delves into the technical details of its development.

"Bringing LLMs into Tynker's dynamic visual block coding ecosystem presented unique challenges," said Tynker CTO Kelvin Chong. "We fine-tuned Llama2 with AI and ML techniques to align with young coders, supporting parallel scripts, visual actors, and kid-centric features such as Physics, AR, sound effects, and Minecraft modding. As coders grow, they're naturally transitioned towards Python.”

Tynker users will be notified when Copilot becomes generally available, the company said.

Learn more at Tynker.com.