Why AI models need long-term support, like Ubuntu LTS · Chris Innes

Why it matters

Chris Innes argues that the AI industry needs a Long Term Support release, modelled on Ubuntu's practice of designating a stable version maintained for roughly five years while development continues elsewhere. Writing in response to Dario Amodei's essay on pacing frontier development, Innes contends that constant model replacement harms organisations like schools, further education colleges and NHS trusts, which build careful workflows, training and safeguarding procedures around a specific model's behaviour. When a model is deprecated or its behaviour shifts, those institutions must redo validation work, and the disruption falls on staff and students who had only just adapted. He distinguishes a support commitment from Anthropic's undertaking to preserve model weights: preserved weights protect the model itself, whereas a support window protects the people who depend on it. An AI LTS would document and version model checkpoints, inference configuration and interfaces, publish compatibility criteria, and maintain regression tests, so that change arrives visibly rather than silently breaking working systems. The piece situates itself within broader debate about whether frontier pace should slow so society can absorb existing capabilities.