Sam Hogan, CEO of Inference, describes training and hosting open models for teams seeking control over cost, latency and task-specific performance. He argues that a model need not be the strongest available if it reliably meets the requirements of a particular workload.
Hogan contrasts relatively focused extraction tasks, where supervised fine-tuning may reduce costs, with harder open-ended research or sovereign-model ambitions. He and Gabriel Dickinson emphasise evaluating production tasks against an organisation’s own scorecards instead of relying only on public benchmarks.
The speakers consider the changing gap between open and proprietary models, possible business strategies for model providers and the effects of AI on knowledge work. Hogan’s cost estimates, market predictions and employment views are presented as opinions rather than guaranteed outcomes.
Watch on YouTube
