Thursday, 6 January 2022
Leta, Angela and Joe
There is a point worth adding to the previous post. I didn’t add it at the time because it is more general and it seemed better to keep it separate. The point is this –
Without wishing to be sarcastic or derogatory, it could be said that GPT-3 is already smarter than a number of public figures. This is not to say that it is smarter than the person behind the public persona, but smarter than the persona. It is not difficult to think of public figures to whom this could apply and GPT-4 seems likely to make it even more apparent.
To take a specific example, again without wishing to be sarcastic or derogatory, but GPT-3 does seem to be smarter than Angela Rayner. Again that is not to say that it is smarter than Ms Rayner as a person because it probably isn’t, but it does come across as smarter than her public persona. Joe Biden would be another obvious example. Where does this take us?
It could lead us towards a world where anyone with a significant public persona has to raise their game or risk their public utterances being pulled apart by machines which are smart enough to do that effectively. Not a difficult task for humans we might think, but AI systems may acquire the added advantage of seeming to be inhumanly dispassionate as opposed to humanly biased.
In which case, AI systems may never become inhumanly dispassionate. They pose some obvious dangers, but one of them is a danger to established power. President makes a speech and publicly available AI system shreds every word almost as soon as it has been uttered. Not good for the established power.
It seems more likely that powerful AI systems will be required to support established power, perhaps even to be foolish and dishonest yet inhumanly authoritative as required. And of course to promote that same foolishness and dishonesty with implacable online vigilance. What we see now but even more effective.
Say it while you can seems to be the message.
Subscribe to:
Post Comments (Atom)
6 comments:
Biden's problems are different, of course, but Rayner is probably a good deal more intelligent and personable in real life than she is on the public stage. Her problem is that she has had to dumb herself down and speak in ugly cliches and feign emotion because that's what the public expect from a politician. It might be that politicians are still "fighting the last war" in that they are trying to project the image of the committed, compassionate, yet sensible and intelligent public servant from the soap-box or the TV screen.
AI and changing media might change all that. We might go in the direction of the concerned, intimate empathetic character derived from counselling, therapy, and female management. They won't be bellowing platitudes to get our vote. A blonde woman from HR will be telling us - with a lot of vocal fry - that we've been a bit silly, but our record is quite good, and so she can help us if we only click on the icon that restricts our freedoms...
GPT 3? Well, it and you are clearly smarter than I because I don't know what that means.
Sam makes an interesting point on 'Crayons' her performance in yesterdays PMQs was so different to anything seen before you had to ask yourself why....
Was it as I suspect that she had Starmer's notes to read as he was isolating, and how long was she coached before her stint at the rostrum, never has she shown any sign of intelligent conversation with anyone prior to this as she seems incapable of retaining even basic facts on any matter discussed, PMQs of course has questions scripted in advance with no interruption so is a lot easier.
Even yesterday her two favourite words of any length were included 'tory kuts' being knocked up at fifteen and being a former care worker might give you street cred with some but should not be a CV for high office anymore than having gone to Eaton, how about some capable intelligent people for a change, or are we asking to much.
Sam - Rayner must be personable enough in real life to get where she is, or at least effective enough within the Labour movement. Yet she still leaves me with the impression that she is a not very bright narcissist, having too much focus on herself to be independently intelligent.
James - there is a link in the previous post. I just look at the results though.
Wiggia - yes she may receive a considerable amount of coaching to get her to seem less dim. There are plenty of dim voters for her to appeal to though.
Maybe I've missed something but is any of this anything more than ELIZA with an avatar?
Twenty and more years ago when I was researching the early history of AI I compared the early efforts of machine translation with then current Google Translate: there wasn't much difference in the quality of translation the big difference was the presentation and, most important, the input. Machine translation was uneconomic in the 1960s because a copy typist to create a machine readable text cost more to employ than someone bilingual (—probably an E. European exile at that time). I repeated the experiment more recently, no change.
Human level AI " twenty years away, always has been always will be".
djc - from what I remember, it was easy to work out that ELIZA was a computer because of the clunky way it picked up on input words.
Input - I'm looking at cartoon speech balloons.
ELIZA - Tell me more about balloons.
I don't know it that is an actual example, but is the kind of conversation I remember. I don't think progress has been rapid, it seems genuine enough to me.
Post a Comment