Companies building entire workflows to replace existing ones on technology that is already expensive, is losing money on every use, failed to show it can reliably do the things it claims, has burned unknown billions in advancing marginally, and will almost certainly get exponentially more expensive as the AI providers realize they have no more investors as we precariously navigate possible global recession and trade/supply chain complications.
Whatever they sell getting adopted, people laid off, actual knowledge dying in companies. And then these companies crashing down when the what was sold does not materialize. And this then cascading elsewhere...
Right, it's just potentially as our more destructive as the thin veneer of legitimacy allows this stuff to steal investment from actually useful things and cause epistemic collapse as it infiltrates our knowledge systems. But at least it's not counterfeit money.
Committing crimes can be useful for humanity. Wherever you live, or come from, I can almost guarantee the current regime in your country was the result of treason, revolution or usurpation. It’s also quite likely that this event was a net positive for your country. Every single rebel, revolutionary and freedom fighter is someone elses criminal and terrorist.
That's hogwash. I've seen in person quite a few successfully implemented LLM backed process automation projects that actually produce better KPIs than the human powered chain that came before.
How many successful, value-add block chain implementation projects have there been? Has there even been a single one?
I've worked on such projects, too. They were carefully selected to have minimum impact when the models are wrong.
In some of these cases, as you say, the KPI was higher, e.g. in classification tasks (where previously a clerk was supposed to label some cases). An LLM produced decent results most of the time - and didn't get tired. But when I looked deeper, it turned out the folks who decided to use LLMs could have used classical ML learning methods as well with similar efficiency.
I believe we arrived to the point where people just got lazy and just use LLM for anything.
This absolutely seems to be the way things are going. We keep hearing about how 'AI' models have done these amazing things and learn they were specially trained models. Most news reports don't even get into if it was an LLM or ML.
Then business leaders think they can just pay for a subscription to an all encompassing LLM model and get the same results. It's just an evolution of the long standing tradition of tech being pushed from the top down because the sales guys impressed the C-suites.
Whats your metric for "Successful, value add" because I get the distinct impression that you are poised to disregard anything from the space.
Also, nothing I have said disagrees with you. I havent said that all AI projects are useless, just identified the cause of the rush. I have seen some toe curling LLM implementations that could only exist thanks to executive stupidity.
Not just AI. Entire science progress is funded by FOMO. Businesses buy some tech only to be competitive. Businesses were happy with their revenue even when there were no computers around. I had to go to moon just because the other stupid guy went there.
"Big Data is like teenage sex: everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it." - Dan Ariely (at least as early as 2018)
What could go wrong?