Brief
New benchmark and method aim to forecast meeting continuations
A preprint describes a benchmark built from 2,207 real meetings and a method called GLARE that generates plausible meeting continuations. The results are from human evaluations, but the work has not yet been peer-reviewed.
The abstract reports that GLARE achieved average human-evaluated win rates of 0.66 on utility and 0.70 on human-likeness, outperforming SFT and SPIN but remaining below human continuations. The benchmark includes 24,794 future-facing queries. As a preprint, these results have not been independently verified.
Our reading
Our reading is that this preprint offers a new way to evaluate meeting forecasting, but the lack of peer review means the win rates should be treated as preliminary.
What to do or watch
Watch for peer review or independent replication before treating GLARE's 0.66 utility and 0.70 human-likeness win rates as settled, since the preprint's results have not been independently verified. The open question is whether those human-evaluated win rates hold up under outside scrutiny, and how the benchmark's arena-style comparisons of general-purpose models are judged.
Source details and supporting facts
Each line is stated by the page named above it.
Stated by arXiv
- The Meeting Dynamic Forecasting Benchmark (MDFB) was constructed from 2,207 real-world meetings and 24,794 future-facing queries.
- GLARE attains average human-evaluated win rates of 0.66 on utility and 0.70 on human-likeness.
- GLARE outperforms SFT and SPIN while remaining below the observed human continuation.
Sources
- arXivText stored 14 September 2026
How this story was checked. Written from the 1 page listed above, stored 14 September 2026; claims checked against that stored text on 14 September 2026.
What that means
- 3 of 3 reported statements were confirmed against the page that carries them; the rest were removed rather than published.
- Figures in the text were required to appear in the stored source text: yes. Identifiers: yes.
- The check reads stored text only: no claim rests on a fresh look that did not happen.
- Where the reporting was silent, the text says so instead of filling the gap.