Friday 10 February 2023

Weekend thread: ChatGTP created C@W

OK, hands up: this blog has been written by ever more sophisticated Predictive Text software since around 2010, culminating triumphantly in its present ChatGTP mode!  Suckers!! 

We've also used plagiarism-detection software on the BTL contributions, and fully half of you people are bots.  Who ever believed "E-K" or "Elby The Berserk" were real people??  (Congrats to *Mr Bill Quango MP*, whose "Virtual Constituency Office ®"  software has been successfully submitting plausible House of Commons expenses claims for over a decade now.)

Here's a great explainer of what's been at work below the surface.  A bit of maths is used in the presentation - but not really required to get the gist.  One or two surprises in there for me, not least the first big point this chap makes - that ChatGTP proceeds only word-by-word.

What else on the www do you reckon must have been bot-generated?  Is Liz Truss a bot from the Ayn Rand Foundation?  And will ChatGTP get swept away by the serious tech giants' competitor products?



Anomalous Cowshed said...

Arrgh! It's a bloody pattern matcher. That's all it is.

It's beginning to look suspiciously like a huge waste of time and cash.

andrew said...

Clearly not there yet
This is just a language model and as such v.v. inefficient.

At some point a better memory model will be added (more efficient)

Later on something that actually learns by itself from self directed observation (this is not correct)

PushingTheBoundaries said...

Buy Google now..

djm said...

Shurely ChatGTP is just another data harvesting operation ?

Caeser Hēméra said...

@ND - point of order, Ayn Rand influenced bots, and their human equivalents, are known as 'Randroids' ;)

With ChatGPT and its ilk, the one thing that doesn't get mentioned too much is cost - each version is a snapshot of the training at a point in time. And it needs retraining as data moves forward. On pricey hardware.

Google's Sparrow will have a better training pipeline, but there is still a cost to keeping it up to date, especially as the inputs move into the trillions and beyond, and Big Tech is starting to listen to the beancounters concerns.

So, unless it drives income or investment, it has a lifespan of its use in killing off the new competition, ChatGPT.

And thats not guaranteed, Google can't move with impunity quite like it once did, with an increasing interest in breaking it up. On the plus side, MS going through similar was the only thing that saved it from a path of obsolescence.

Nick Drew said...

Anon @ 1:58 - gotta love " In recent years, gas prices in the UK have been relatively stable" !

Other than that ...

dearieme said...

I used to work round the corner from a university department of AI, or "machine intelligence" as it was known at the time.

The departmental divas boasted long and loud about their imminent glorious success. But I had a beer with one of the younger staff who worked at the codeface. His tale was quite the opposite, summarisable as "we don't know what we're doing".

So, have things changed in a few decades or is it like fusion power plants, doomed always to be the future?

Don Cox said...

Was the "machine intelligence" department in Edinburgh ?

I worked in a department called Machine Intelligence and Perception around 1969. I was more on the perception side. The MI folk seemed to be having fun.

Back then I had never written a line of code, so I couldn't join in.


andrew said...

From memory, the best places 20 years ago were Manchester Edinburgh and Warwick. I did a postgrad in ai at Brunel in the early 90s part time. I liked their fairly dismissive attitude to cleverness, they just wanted to solve 'simple' stuff like good or applying fits to heart monitors

andrew said...

Oops good OCR and FFTs
So much for ai in spellcheck

Lord T said...

ML was the start of AI. However, back then we didn't have to hardware, either processing clout and disk space. Things have moved on and now the hardware is here, enough for this stage with better coming, so the bright boys can make it work.

It's here, its still at the beginning but its already showing great promise and, unfortunately, already showing how these people are hampering it with wokeness.

Anonymous said...

As you can see from the "ND essay", it's good enough for school essays, and these days probably for uni ones too. People are already using it.

My understanding is that unis already have some kind of anti-plagiarism software in place, they'll need to add these new AI platforms to them.

(the secret IIRC is to reshuffle paragraphs and do a bit of rephrasing... its a sort of arms race twixt students and lecturers, 'we pretend to teach them and they pretend to learn, firsts all round'.)

It is though actually quite useful for an old IT guy wanting to develop coding skills. You can ask it to build something simple in say python for you and then you can take things forward yourself.

PushingTheBoundaries said...

@dearieme They are now at the stage where some AI models are generating results that the programmers cannot understand how they are generated. This is both a blessing and a curse. A horizon of unknown opportunities awaits, or a descent into potential human decline.. who knows eh!

Caeser Hēméra said...

@dearieme - we're nowhere near actual AI, and unlikely to be so for some time, decades, maybe even centuries.

What we have now are models are that are very good for defined criteria, married to the usual marketing to get more funds in, and once outside of those boundaries they're less use than a village idiot attempting to install chocolate fire cladding on a high-rise.

We have cancer tracking models, which are impressively accurate - just so long as the scan isn't off-angle, then they're frighteningly inaccurate. Hands up who wants their cancer missed because the intern plopped the scan in a single degree off-kilter?

We also have issue of asking the right questions - one ML was asked to beat another at noughts and crosses on an infinite board. It ascertained the quickest way to win was to put a marker at such a distance the opponent crashed.

And yes, some are now so complex, we don't fully know how it got to the answer. Not knowing how something works is not a marker for intelligence though, anyone who has worked in an office ought to be able to attest to that.

ML is going to change a lot, but we need understand the gap between what we say and what we mean, it has no contextual capability for that. Ask for the house to be made spotless, don't be shocked if it skins your teenagers...

dearieme said...

@Don; yup.

Anonymous said...

Is the Daily Mail written by malfunctioning ChatGPT bot which has just got past the word Diana and is stuck in an endless loop at the word Meghan?

andrew said...

Perhaps someone should ask chatgpt?

E-K said...

Well that's it. Caught red handed...

I'm a botty man !

Anonymous said...

Meanwhile in China:

"Things are escalating again in Ukraine, we must distract America with a putative alternate enemy threat; a substitute expensive military adventure. But we don’t want to do anything to spook these idiots too badly. Anyone? General Wang of the People’s Balloon Ministry, your hand is up… do you have an idea? “

PushingTheBoundaries said...

Yes, all very good points especially wrt asking the right question(s)

Anonymous said...

OT but someone on MoA just posted Putin's 2021 speech referencing the attempted coup in Belarus (how soon we forget). I had no idea he was a Kipling fan.

"We want to have good relations and good communication with everyone But we see what happens in real life: As I said, Russia is being attacked here and there for no reason at all. And of course, just like Sher Khan, all sorts of little Tabaquis are circling round them (the US), all howling in order to appease their sovereign. Kipling was a great writer."

He had and has a point. I'm particularly worried about the Poles, whose government I agree with on just about everything except their insane hatred of Russia, no matter how explicable by history. Churchill's words are sadly still true:

“The heroic characteristics of the Polish race must not blind us to their errors, which over centuries have led them through measureless suffering…it is a mystery and tragedy of European history that a people capable of every heroic virtue, gifted, valiant, charming, as individuals, should repeatedly show such inveterate faults in almost every aspect of their governmental life.”