Electricity was world-changing, do we consider ourselves to be using it? Why is there this presupposition with LLMs? Are we just taking the "ai" leaders for their word? From what I can see, the tech we have today is not up to to the task of replacement and capability improvements (not benchmark) seem to be decelerating.

▲

almostdeadguy 6 hours ago | parent [-]

> Electricity was world-changing, do we consider ourselves to be using it?

We do, but craft workers in the industrial revolution did lose their jobs, and at the pace AI is developing I don't see any reason to believe that professional workers will not eventually have their jobs threatened.

> Why is there this presupposition with LLMs? Are we just taking the "ai" leaders for their word?

No, I don't think we are. And let me be clear: I was very skeptical about this until recently. I think the models are objectively capable of doing a lot of the job of software engineers now, and increasingly engineers are being pressured to not only use them, but to take themselves out of the loop. There is a reason the conversation of "should we read the code" has been raised in public discourse and engineering talks now. That should concern everyone.

> From what I can see, the tech we have today is not up to to the task of replacement and capability improvements (not benchmark) seem to be decelerating.

I am not seeing that: https://metr.org/time-horizons/

And I think if things continue as they have, we are eventually going to see the price of software decline and widespread layoffs soon to follow.

	▲	verdverm 3 hours ago \| parent [-]
		Jobs were not just lost during the industrial revolution, how many new tasks and jobs were created? (hint, it was more than were lost) If someone is telling you to not be in the loop, they are not aware of the capability ceiling of agents and llms. Progress appears to be slowing the last few months, people are not finding Opus 5 to be that much better than 4.8, less wow effect these days. To assume that Ai will lead to mass layoffs is to predict an uncertain future. I'm not worried about it this after years of using these LLMs and agents. There was a capability point reached late last year, the improvements since are minor by comparison. I've had sufficient interactions with others at work (using Ai on code I wrote) to know they are not up to the task. Their attention is too narrow, their fixes locally optimal. They still make stuff up all the time just to be pleasing.