Friday 31 May 2024
SLM
The other day found me chatting about AI developments with my mate Dr Baz Broxtowe of Fradley University. We met up at a service area on the M1 motorway because Dr Baz was on his way to an artificial intelligence conference up north somewhere.
It was noisy because lots of kids were obviously off school for half-term or something, but we found a table where Dr Baz was able to outline some recent AI developments in the political field.
“As you probably know, it’s all about language,” said Dr Baz when I finally managed to grab a couple of coffees and work my way back to the table without scalding one of the kids rushing around the place.
“AI systems are essentially Large Language Models or LLMs,” he continued, somehow managing to raise his voice over a nearby argument about doughnuts. “They use complex mathematical algorithms to learn appropriate linguistic responses by being trained on vast inputs of language already in use.”
“Okay I get that,” I replied, “apart from the maths and computing involved I can see how it works in a very, very general sense.”
“In that case you will understand why political parties are interested in it,” Dr Baz continued.
“Boost the apparent intellectual quality of their candidates?”
“Essentially yes. They are interested in the possibility of a kind of dynamic autocue where an AI system is able to formulate responses dynamically even when picking up verbal questions from a live audience. These responses would be piped to the candidate through an autocue.”
“What about when they are out and about canvassing and so on?”
“This is the interesting aspect, the AI system would pick up, say a journalist’s question or maybe a question from the public and it would pipe a response into the candidates earpiece.”
“To a lay person that sounds tricky,” I replied.
“It is tricky, and that’s why the Greens have gone for a different approach.”
“How different?”
“Instead of AI based on a Large Language Model, they have gone for a Small Language Model or SLM as we call it. As they say much the same thing to any question, the Greens think an SLM will do well enough. Really it’s just a collection of what we call eco-phrases, but they seem happy with it.”
“Interesting.”
“It is, isn’t it?” Dr Baz replied as he sipped his coffee.
Subscribe to:
Post Comments (Atom)
4 comments:
The question is whether, or not, you are willing to be persuaded by political debate which consists of distilled parables, aphorisms, slogans and gossip?
We already do live in such a world of course, but the AI Language Models will make it more immediate and 'catchy'.
Perhaps we should let the politicians build a Red LLM and a Blue LLM (other colours are available) and allow them to have a public debate? But I rather suspect the 'result' will be a soggy middle - which we have already achieved by natural means.
Fleabay has an option of using "AI" to describe what you're selling: the results are awful. They read as contrived, repetitive and - not surprisingly - artificial. As expected, lots of A and not enough I.
DJ - it's a grim world where adults allow themselves to be persuaded by those distilled parables, aphorisms, slogans and the gossip. Yet maybe it has to be like that and all AI will do is to reinforce what has to be, there has to be consensus around undemanding language. Charlatans are good at undemanding language, always will be unfortunately.
Jannie - Amazon has an AI description of items sold through it and apparently this is taken from reviews. From what I've seen it does work, but it is still better to look at the reviews, those that seem to be genuine anyway.
Post a Comment