2026 · 09 · 162026-09-16September 16, 2026
By Florent Tavernier · Sundial ResearchBy Florent Tavernier, Sundial Research
How we fine-tuned Inkling-Small on the largest dataset of real LaTeX compilation errors to date and got it ahead of every API model we tested, in under a second per fix.
2026 · 09 · 092026-09-09September 9, 2026
By Matthew Diakonov and Belinda Mo · Sundial ResearchBy Matthew Diakonov and Belinda Mo, Sundial Research
We added Lean 4 agent workflows to a LaTeX editor so that researchers can use and understand auto-formalizations as they write their papers. While a researcher writes, agents translate their work into Lean, and the compiler decides what is proved, open, or broken.
2026 · 06 · 192026-06-19June 19, 2026
By Belinda Mo and Florent Tavernier
Increasingly, people can't explain what their agents did for them. We built Sundial so humans stay at the center of the work: a collaborative workspace where humans and agents edit the same files live, and every change can be reviewed at an exact level of granularity.
2026 · 06 · 142026-06-14June 14, 2026
By Belinda Mo · ICML 2026 position paperBy Belinda Mo, ICML 2026 position paper
AI systems are becoming autonomous research agents, and the verification gap between scientific output and our ability to check it is widening. We argue science must evolve its verification infrastructure: observable-by-default workflows, tiered verification, and clear attribution.