This afternoon, I idly sketched out ‘there’s no such thing as six-fingered text’.
Some of my work of late has taken a sudden AI twist. I mean, whose hasn’t… in the discursive, theoretical stage at least. Lots of people talking about it, nobody really doing it.
And with good reason, perhaps. As conversations unfold, you realise just how uncertain the results of embedding LLMs in various regards could be – wonderful on one hand, woeful on the other. Loads of possible upsides, loads of downsides. An exciting time to experiment, a cautious time to commit.
Which brings me back to ‘six-fingered text’.
Much was made a while ago of the inability of visual GenAI to create accurate hands. Part of the tell-tale signs of a faked AI image was how many fingers a hand would have. Then it became big news when the hands problem was sorted.
The thing about text passages, however, is there is no six-finger equivalent. No easy glance that can tell you what’s factually incorrect, as opposed to visibly disturbing.
And if we’re less predisposed to spot these issues, are AI companies less likely to fix them as quickly?