Coding models are running out of data — PL researchers propose 'intent computing' as the fix
LingmingZhang · x · 2026-09-03
Grigore Rosu warns that LLM coding ability depends heavily on code data, and the industry has essentially consumed all high-quality code data for training and evaluation — progress is visibly slowing. Whoever finds the next big data pool holds the key.
His group, rooted in programming languages and formal methods, proposes a new paradigm called intent computing:
- Code should ship with its specification and a machine-checked proof;
- This eliminates the need for humans to spend time reviewing the code itself, changing what an AI answer is;
- The team has run several experiments, with more details to come.
The core idea: formally verified code-plus-proof artifacts could be the next major pool of high-quality training data.
More from coding & agent
- Anthropic's new Claude Fable 5.1 docs: one prompt line removes 'Claude-speak' — daniel_mac8 · 2026-09-03
- Robot duck learns to skateboard on a Blender-MCP-designed 3D-printable board — TinfoilTricorn · 2026-09-03
- NVIDIA open-sources a tool that scans AI agent skills for security risks before you run them — Roger_M_Taylor · 2026-09-03
- Coding agents can already act like recursive language models, says Alex Zhang — CShorten30 · 2026-09-03
- Stanford launches CS329Z, a new fall course on engineering AI agents from scratch — Diyi_Yang · 2026-09-03
- Calling Gemini Managed Agents from Google Apps Script, streaming artifacts to Drive — rseroter · 2026-09-03