Post

Matthew Adams
@ythos.bsky.social

AI is creating waves by automating bad decision making, poor quality writing, slapdash research, and mediocre art. The fact it is perceived as revolutionary tells us a lot about our terrible middle managers and their slapdash decision making, poor quality research, and desire to make bad "art".

15 reposts60 likes

Matthew Adams
@ythos.bsky.social

I'm still hopeful that the planet burning AI that has enshittified so many tools I interact with every day (Google/Office/customer service interactions) will give way to a next generation of small, useful agents that don't try to do more than they are capable of.

Matthew Adams
@ythos.bsky.social

It will require a careful dance to avoid terrifying investors. No-one will ever be allowed to admit that we've been wasting billions of dollars and burning the planet because we bought into a fantasy. So we are gradually changing the frame of that fantasy and moving it towards a useful reality.

Frustrert uczeń
@sfrustrowanyelev.bsky.social

I hope it's a bit like the dot-com bubble around 2000, where we had similar outrageous claims of what the internet could do, followed by a quick crash, but then some actually useful internet services like classic Google, Wikipedia, non-rapacious Amazon, etc.

Matthew Adams
@ythos.bsky.social

On the other hand there's great work automating small repetitive tasks using traditional ML techniques that are getting better and faster. And small model language processing to provide a more natural interface to those very specific tools, and to allow tools to talk to one another.

Piersy
@mockneypiers.bsky.social

This is it, I keep getting people in the co. saying "how can we use AI" and I generally reply "Don't". If they say "can we automate processing photos of cow manure for signs of mycotoxin" then there's ML near you that can help.

Juliet E McKenna
@julietemckenna.bsky.social

A journalist I know had been using a transcription program along those lines for months, to save time typing up phone interviews, said it was 90+% accurate, highlighting 'unintelligible', very useful. New release with 'added AI' was useless, inserting predictive-text sentences that no one had said.