Is it OK to say "slop" again? Microsoft boss Satya Nadella took to the stage on the London leg of the company's AI tour and said the words that many an IT pro has uttered when faced with a Copilot rollout: "Nobody wants anything that is sloppy in terms of AI creation."
No, they do not. Nadella was talking about AI assistants, agentic AI, augmenting work, and ensuring that the next person in the data chain understands how the output was produced. However, the CEO of Microsoft dropping the word "sloppy" following his well-publicized request that we all move on from denigrating the output of AI is certainly an eyebrow-raiser.
Microsoft's AI tour – aside from some awkward scalability issues at London's Excel (if you saw the queue for the badge collection, you'll know what we mean) – was unsurprisingly all about the company's ambitions for AI. Copilot featured large, and so did the "infinite set of minds" (Nadella's words) afforded by AI-powered agents. But the elephant in the keynote auditorium – almost omnipresent on the screens but not directly addressed – was the fact AI output cannot be trusted.
Satya Nadella delivering the keynote for Microsoft's London AI tour
For every whizzbang demonstration showing AI tools collating data in Excel, or creating and executing test plans for websites, there was a message warning that the output of AI tools can't be entirely trusted, and needs human verification.
Even a command-line demonstration had the warning: "Copilot uses AI. Check for mistakes."
In a conference heavy on the joys of AI, the on-screen warnings and reminders highlight that AI is far from infallible.
Although the conference leaned heavily into UK AI use cases – including a doctor describing time savings in patient interactions and the oft-cited 26-minute statistic for civil servants – Microsoft avoided mentioning West Midlands Police's Copilot mishap, in which the tool hallucinated a football match. The force's Chief Constable, Craig Guildford, later took early retirement.
But as the keynote screens made clear, AI output cannot be trusted, and "nobody wants anything that is sloppy." That's especially true when making policing decisions or calculating the capacity of a conference center. ®
Source: The register