Ministers’ favourite chatbot has been given elocution classes to stop it spouting ‘trash’ – and get it to talk ‘rubbish’ instead

(Image: AP)

Whitehall’s AI assistant has been given elocution lessons to stop spouting ‘trash’ and talk ‘rubbish’ instead.

The AI tool set, nicknamed “Humphrey” after the manipulative civil servant in TV’s Yes, Minister, has been used to cut back on expensive consultants and speed up how government departments operate.

But users noticed a flaw in the bot – an irritating tendency of using Americanisms.

Keen to make sure ministers’ favourite official remained a true Brit, AI experts in Whitehall’s Technology Ministry built a translator for Humphrey, known informally as “uwotm8”.

Technology Secretary Peter Kyle said: “Humphrey has the potential to transform the way government works – making things faster, more efficient, and less reliant on expensive consultants as we create a leaner state focused on delivering Plan for Change priorities.

“But an AI tool named after a British sitcom icon must speak the King’s English. With this new translator, he now sounds a bit more like the rest of us – and that matters when he’s advising ministers or engaging with the public. It’s a simple fix with a big impact.”

The news comes as Elon Musk revealed a chilling plan to re-write history using his AI chatbot – with readers accusing him of copying 1984.

In George Orwell’s dystopian novel, hero Winston Smith works at the Ministry of Truth, where his job is to rewrite historical documents and newspapers so they match the tyrannical government’s constantly changing party line.

This morning, Musk vowed on X to use the latest version of AI chatbot Grok to “rewrite the entire corpus of human knowledge, adding missing information and deleting errors.”

He said Grok, which X users can access directly within the app, would be “retrained” based on the “corrected” data.

READ MORE: Join our Mirror politics WhatsApp group to get the latest updates from Westminster

AI systems are trained on huge sets of data – mostly from publicly available sources like books, newspaper articles and other sources on the internet.

ChatGPT, the main competitor for Musk’s Grok AI, is estimated to be trained on more than a trillion words of information.

Musk’s suggestion would be for his next model to be trained not on original historical sources, but on Grok’s revisions of them – with the erratic tech billionaire’s team stepping in to remove “errors”.

Musk posted: “Far too much garbage in any foundation model trained on uncorrected data.”

Share.
Exit mobile version