Turkey earthquake response shows why banning open models can slow critical AI work

mervenoyann · x · 2026-07-23

The author argues against banning open models, using a 2023 Turkey earthquake response as an example of why model access and vendor lock-in matter. They needed large-scale entity and sentiment classification on Twitter to help locate survivors, but GPT-3.5 was too expensive and the team waited days for a reply. In the meantime, they trained BERT/RoBERTa and used Hugging Face’s inference API for free.

The point, they say, is not that closed models are bad — they like using them — but that AI systems are becoming critical infrastructure, so restricting open models can slow down urgent work. The post ends with a blunt defense of open access: “own your intelligence.”

Original post →

More from AGI Musings

AGI Musings channel →