AI Exchange @ AWS

Adel made the case for small language models: lightweight enough to run on a laptop, yet still strong enough for real work, and cheaper, more private and more practical for business use than reaching for the largest model available. Quantisation, distillation and fine-tuning were the techniques that get them there.

His framing was the one that actually lands in a board meeting — why your CFO should care. Data stays home, which satisfies GDPR, local regulation and internal policy in one move. Smaller models are easier to interpret and audit, and narrowing the task flattens the hallucination surface. And the cost is dramatically lower, with no surprise pricing, no rate limits and nobody asking why the API is down at three in the morning.

Elliott, from our hosts at AWS, put the productivity case and its cost side by side. AI measurably improves both output and quality — and over-reliance on it, through plain cognitive offloading, can quietly weaken the skill it was standing in for.

He was open about his own experience of losing an edge he used to have, set it against research on what these tools do to thinking and memory, and made the point that personality shapes how people reach for AI in the first place — so the risk is not evenly spread across a team.