All articles
Technology
7 min read
AI Without the Hype
Three days of main-stage talks, distilled into the claims that actually held up under questioning.
Words by Priya Raman
Every conference has a moment where the vocabulary outruns the substance. Convoke had fewer of those than most, and the reason was structural: every main-stage claim had to survive fifteen minutes of open questioning.
What held up
The strongest sessions were the least theatrical. Teams shipping models into regulated industries talked about evaluation harnesses, latency budgets, and the unglamorous work of getting a system to fail predictably. Nobody used the word “transformative” without being asked to define it.
A recurring theme: the hardest part is rarely the model. It is the interface around it — the affordances that tell a person how much to trust what they are looking at, and the escape hatch for when the answer is wrong.
What did not
Roadmaps presented as inevitabilities. Benchmarks without baselines. Any demo that could not be interrupted. The audience was generous but not credulous, and that combination produced better talks by the third day.