Engine Internals Power Ceiling

[ThunderPigeon]

At my work the company has been really pushing AI (mind you, I'm an engineer) and anyone who parrots the company pro-ai views gets praise even when the more skilled technical workers dislike it. Had a coworker the other day who's been at the job for less than a year rant in a pretty big team meeting about how he starts his day by turning on ChatGPT. Said he thinks the future was to replace learning from your peers with ai learning. Kinda made me really angry.

It is all corporations these days and the worthless corporate simps are always bringing up AI at meetings. I see it as more of a "look at me, I'm up on the new thing" approach to kiss their ass up the corporate ladder.

There is an old saying in the programming world. Garbage in, garbage out. That is the biggest problem with AI. When you look at where most of the information is sourced from, places like Reddit are on the top of the list. That tells you all you need to know about the accuracy of AI.

Sponsored

 

[zumbooruk]

Cognitive offloading to AI is a concern (MIT Study). Mytopia, dopamine, reduction in critical thinking skills over time are additional concerns.

If those are not checked, then IMHO it leads to smart people no longer willing to share their knowledge. For AI to be helpful, it needs to be locally trained from a reliable/trusted dataset and not simply scraping subjective content posted on the Internet. I love reading engineering research/content posted by experts in their field but fear that their content will be diluted or simply no longer shared.

https://ideas.repec.org/p/cpr/ceprdp/21577.html

Generative AI use by Chinese secondary students produces clear short-term homework gains yet generates substantial learning losses on independent tests. Monthly closed-book exam scores fall by about 20 percent within six months. High-stakes entrance exam scores decline by 18 to 24 percent after roughly two years. Losses concentrate among the 80 percent of users who outsource homework to AI rather than using AI as a supplement. High-achieving students, boys, younger students, and social-science subjects suffer the largest penalties.

 

[zumbooruk]

My thesis from 1994 was about Neural Networks. Wrote my own backprop code from scratch. Tiny compared to today's networks, and was used for classification, not generative.

Back then I called it "definitely artificial and definitely not intelligent". I still stand by this today.

Generative AI is a statistical model that predicts the next word from patterns in data.
It will never have any true knowledge.
It will never be able to reason or think.
It will never have any emotions.
It can only copy surface forms of these traits so well that the copy looks indistinguishable from the "real thing"

These systems learn by adjusting billions of numbers so that the most probable next word or phrase follows the previous ones. The process is pure probability. No symbols are grounded in the world. No internal model tracks cause and effect. The output is fluent text that matches human writing habits, yet often time partially or totally wrong...

"don't trust, and verify"

 

[J_D]

You are forgetting the Achilles heel. It’s the most power they could extract while meeting emissions regulations.

Same reason the high revving K’s of the past are in history, they don’t burn clean

Sponsored