MTP: A Meaning-Typed Language Abstraction for AI-Integrated Programming
summary
Presents MTP, a language-level way to call LLMs that adds a by operator and a runtime pipeline that extracts code semantics into MT-IR and drives an automated MT-Runtime for prompt generation, inference, validation and correction. Implemented in Jac as the byLLM module (Jaseci open source), it claims 3.2× faster developer task completion, 45% fewer lines in a user study, ~10.7× fewer integration lines in benchmarks, and robustness when up to 50% of names are degraded. Artifacts and reproduction data are available.
AI-written summary · about
chapters (13)
- 0:00MTP research project intro
- 1:28Game level generation example
- 2:53Prompt engineering limits
- 3:56Meaning type programming hypothesis
- 5:19The by keyword abstraction
- 6:23Meaning type intermediate representation
- 7:26Semantic registry and context extraction
- 8:53Prompt generation and LLM inference
- 10:17Implementation and evaluation setup
- 11:41Developer complexity reduction results
- 12:26Accuracy across LLM models
- 12:46Contributions and ecosystem summary
- 13:27Q&A
related talks
19mBAML: a new programming language for AI
34mLLMs, Meet Your Tools: The Magic of Model Context Protocol
44mBeyond the Prompt: How MCP Connects Language Models to Real Tools
12mMCP and the Dawn of Agentic Intelligence
43mLLMs, Meet Your Tools: The Magic of Model Context Protocol
45mWhere AI Meets Code
29mMCPs are Boring (or: Why we are losing the Sparkle of LLMs)
25mFrom Idea to Integration: An Intro to the Model Context Protocol (MCP)