GOV.UK chatbot gets smarter but slower as LLMs improve

More powerful large language models (LLMs) are helping make the UK government's in-development chatbot more accurate but are also slowing it down, according to the Government Digital Service (GDS). GDS has run two public pilots of its GOV.UK Chat service, the first on a few pages of the GOV.UK website in late 2024 and the …

[cyberdemon]

Just employ humans, ffs

The amount of energy and money wasted on these bullshit generators, in the middle beginning of an energy crisis is just obscene..

We have finite gas, and finite electricity generation/transmission/distribution capacity. We should use what we have on building houses and keeping them warm, not spaffing our energy supplies on layers and layers of unsustainable slop

"It'll create jobs!" ... Bollocks will it. Datacentres are pretty much devoid of all life.

Spend the money on humans, save the energy

[elsergiovolador]

Re: Just employ humans, ffs

Datacentres are also built on foreign technology and are typically foreign owned. This means they'll seek to take out of the UK economy more than they put in.

Unfortunately as it seems government cannot grasp the basics of economy, or maybe the basics are hidden under a pile of brown envelopes.

[Anonymous Coward]

Re: Just employ humans, ffs

Have you met the average civil servant?

Signed, a civil servant!

[ForthIsNotDead]

Hang on...

What happens if you chat with a Govbot and it gives you some bullshit info that is totally wrong, and you get in to trouble for it?

Bot: "Hi there! Welcome to HMRC. I'm a chatbot and I'm here to help you. How can I help you today?"

Meat bag: "Hello - I was wondering if I could get an extension on my tax return filing date - there's additional information I need to send and it will take some time."

Bot: "I've checked your past tax return history and I see that you have an excellent history of filing accurately and on time. Are there any significant changes that you expect to report in this year's return?"

Meat bag: "Not really - but I've traded some crypto, and cashed in some shares and I need to make sure that is taken into account."

Bot: "Well, given your history, we can use your previous year's tax return for this year, and give you a six month extension to make amendments. Does that sound okay?"

Meat bag: "Oh yes, that's perfect, thanks!"

Bot: "No problem. I'm taking action to ensure your previous tax return is used as a basis for this year, and extending your submission date out by 6 months. Is there anything else I can help you with?"

Meat bag: "No that's great, thanks"

Bot: "Thank you for using the HMRC chat bot service. Have a great day!"

6 weeks later, by mail:

YOU HAVE BEEN FINED £1000 FOR NOT FILING YOUR TAX RETURN YOU PLEB.

Now what? If it turns out the bot was just spouting nonsense, who's in the wrong? Will HMRC stand by their bot?

[ACZ]

Re: Hang on...

But it won't do that. There's no way that it will do anything for/on behalf of anybody - odds are that it will use the existing curated .gov.uk website content to provide information (not do anything), citing it's sources.

One big problem is that the gov.uk guides/manuals (e.g. HMRC stuff) are all very carefully written and reviewed, but address really complex issues and so can be long and difficult to navigate/understand. If they can get it to use that information to provide clear and easy to understand answers to user questions then that would be great.

[cyberdemon]

Re: Hang on...

I think you've missed the OP's point.. Of course it won't do anything for/on behalf of anybody.. It's just an LLM. But as an LLM given a "customer service" context, it will pretend to provide services that it cannot provide, and will pretend to take actions which it cannot take, because it is parroting data that it or its base model was trained on..

This has been a common problem with bank chatbots, where they assure the user that some service has been completed, and in fact nothing of the sort has happened.

So, someone could indeed be convinced by a chatbot that they have sent the relevant data to HMRC, and then the HMRC's "real computer", having not actually been updated with said data, proceeds to fine the person for not filing the information.

[Fred Dibnah]

Re: Hang on...

A chatbot accuracy where 1 in 10 queries are answered incorrectly is completely unacceptable when we are dealing with official information.

[katrinab]

Re: Hang on...

The correct answer is of course:

You put in your best estimate of what you think your taxable income is.

You tick box 20 on page 8 of the return to say that it contains provisional figures.

In box 19 on page 7, you explain the situation.

[elsergiovolador]

Negligence

We are now using Amazon’s Bedrock platform and Anthropic’s Claude models to power the latest version of GOV.UK Chat.

Maybe they are not aware that these are covered by US Cloud Act and it means hostile US government has access to all the data if they wish so, without UK government ever knowing about this.

This means any queries you make are not private between you and UK government.

Typical level of incompetence and negligence.

[Andy The Hat]

Re: Negligence

^^this, simply this.